U.S. patent application number 15/590715 was filed with the patent office on 2017-08-24 for gap measurement for vehicle convoying.
The applicant listed for this patent is Peloton Technology, Inc.. Invention is credited to Stephen M. ERLIEN, John L. JACOBS, Stephan PLEINES, Austin B. SCHUH, Joshua P. SWITKES.
Application Number | 20170242443 15/590715 |
Document ID | / |
Family ID | 59629906 |
Filed Date | 2017-08-24 |
United States Patent
Application |
20170242443 |
Kind Code |
A1 |
SCHUH; Austin B. ; et
al. |
August 24, 2017 |
GAP MEASUREMENT FOR VEHICLE CONVOYING
Abstract
A variety of methods, controllers and algorithms are described
for identifying the back of a particular vehicle (e.g., a platoon
partner) in a set of distance measurement scenes and/or for
tracking the back of such a vehicle. The described techniques can
be used in conjunction with a variety of different distance
measuring technologies including radar, LIDAR, camera based
distance measuring units and others. The described approaches are
well suited for use in vehicle platooning and/or vehicle convoying
systems including tractor-trailer truck platooning
applications.
Inventors: |
SCHUH; Austin B.; (Los
Altos, CA) ; ERLIEN; Stephen M.; (Mountain View,
CA) ; PLEINES; Stephan; (San Mateo, CA) ;
JACOBS; John L.; (Bonny Doon, CA) ; SWITKES; Joshua
P.; (Mountain View, CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Peloton Technology, Inc. |
Mountain View |
CA |
US |
|
|
Family ID: |
59629906 |
Appl. No.: |
15/590715 |
Filed: |
May 9, 2017 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
PCT/US2016/060167 |
Nov 2, 2016 |
|
|
|
15590715 |
|
|
|
|
62249898 |
Nov 2, 2015 |
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06K 9/00791 20130101;
G05D 1/0257 20130101; H01Q 1/3233 20130101; G01S 2013/932 20200101;
G01S 13/865 20130101; G01S 2013/9318 20200101; G01S 2013/9325
20130101; G01S 2013/93271 20200101; G01S 13/867 20130101; G05D
1/0293 20130101; G05D 2201/0213 20130101; G01S 17/931 20200101;
B60W 30/00 20130101; G01S 13/931 20130101; G01S 2013/9316 20200101;
G01S 2013/9319 20200101; G08G 1/22 20130101; G01S 19/14 20130101;
G01S 2013/93185 20200101 |
International
Class: |
G05D 1/02 20060101
G05D001/02; H01Q 1/32 20060101 H01Q001/32; G01S 13/93 20060101
G01S013/93; G01S 19/14 20060101 G01S019/14; G06T 7/20 20060101
G06T007/20; G06K 9/62 20060101 G06K009/62 |
Claims
1. A method of identifying a position of a back of a first vehicle
using radar scenes received from a radar unit on a second vehicle,
the comprising: a) estimating a position of the first vehicle
relative to a second vehicle; b) receiving a radar scene sample
from the radar unit on the second vehicle, the radar scene
including a set of zero or more detected radar object points, each
radar object point corresponding to a detected object; c)
identifying first vehicle radar point candidates within the set of
received detected radar object points; d) categorizing the first
vehicle radar point candidates based on distance that the detected
objects that they represent are from the estimated first vehicle
position; e) repeating steps (a)-(d) a multiplicity of times,
whereby the categorized first vehicle radar point candidates
include candidates from multiple sequential radar scene samples;
and f) identifying the position of the back of the first vehicle
based at least in part of the categorization of the first vehicle
radar point candidates.
2. A method as recited in claim 1 further comprising identifying a
bounding box around the estimated position of the first vehicle,
wherein radar object points within the set of received detected
radar object points that are not located within the bounding box
are not considered first vehicle radar point candidates.
3. A method as recited in claim 2 wherein the bounding box defines
a region that exceeds a maximum expected size of the first
vehicle.
4. A method as recited in claim 1 further comprising estimating a
speed of the first vehicle relative to the second vehicle, the
estimated relative speed having an associated speed uncertainty,
wherein radar object points within the set of detected radar object
points that correspond to detected objects that are moving at a
relative speed that is not within the speed uncertainty of the
estimated speed are not considered first vehicle radar point
candidates.
5. A method as recited in claim 1 wherein the identified back of
the first vehicle or an effective vehicle length that is determined
based at least in part on the identified back of the first vehicle
is used in the control of the second vehicle.
6. A method as recited in claim 1 wherein steps (a)-(c) are
repeated at a sample rate of at least 10 Hertz.
7. A method as recited in claim 1 wherein categorizing the first
vehicle radar point candidates includes populating a histogram with
the first vehicle radar point candidates, the histogram including a
plurality of bins, each bin representing a longitudinal distance
range relative to the estimated position of the first vehicle.
8. A method as recited in claim 7 wherein the identification of the
back of the first vehicle is only done after the histogram contains
at least a predetermined number of first vehicle radar point
candidates.
9. A method as recited in claim 7 further comprising applying a
clustering algorithm to the first vehicle radar point candidates to
identify one or more clusters of first vehicle radar point
candidates.
10. A method as recited in claim 9 wherein the clustering algorithm
is a modified mean shift algorithm.
11. A method as recited in claim 9 wherein the cluster located
closest to the second vehicle is selected to represent the back of
the first vehicle.
12. A method as recited in claim 9 wherein the cluster located
closest to the second vehicle that includes at least a
predetermined threshold percentage or number of first vehicle radar
point candidates is selected to represent the back of the first
vehicle.
13. A method as recited in claim 12 wherein the predetermined
threshold percentage is at least 10% of first vehicle radar point
candidates in the histogram.
14. A method as recited in claim 12 wherein the predetermined
number of first vehicle radar point candidates is a number that is
at least 40.
15. A method as recited in claim 1 further comprising determining
an effective length of the first vehicle based at least in part on
the identified back of the vehicle.
16. A method as recited in claim 1 wherein Kalman filtering is used
to estimate the position of the first vehicle.
17. A method as recited in claim 7 further comprising comparing
properties of the histogram or mean shift clusters derived from the
histogram to a known set of data representative of a target partner
vehicle to verify whether the first vehicle is the target partner
vehicle.
18. A method as recited in claim 7 further comprising comparing
properties of the histogram or mean shift clusters derived from the
histogram to a radar scene received when the back of the first
vehicle is not within the radar unit's field of view but a portion
of the first vehicle is within the radar unit's field of view to
help determine a current relative position of the first
vehicle.
19. A method as recited in claim 1 wherein the first and second
vehicles are trucks.
20. A method as recited in claim 19 wherein the first vehicle is a
tractor-trailer truck.
21. A method of identifying a position of a back of a first vehicle
using scenes received from a distance measuring unit on a second
vehicle, the comprising: a) estimating a position of the first
vehicle relative to a second vehicle; b) receiving a scene sample
from the distance measuring unit on the second vehicle, the scene
including a set of zero or more detected object points, each object
point corresponding to a detected object; c) identifying first
vehicle point candidates within the set of received detected object
points; d) categorizing the first vehicle point candidates based on
distance that the detected objects that they represent are from the
estimated first vehicle position; e) repeating steps (a)-(d) a
multiplicity of times, whereby the categorized first vehicle point
candidates include candidates from multiple sequential distance
measuring unit scene samples; and f) identifying the position of
the back of the first vehicle based at least in part of the
categorization of the first vehicle point candidates.
22. A method of tracking a specific lead vehicle using a distance
measurement unit mounted on a trailing vehicle, the method
comprising: (a) obtaining a current sample from the distance
measurement unit, the current sample including a set of zero or
more object points; (b) obtaining a current estimate of a state of
the lead vehicle corresponding to the current sample, wherein the
current estimate of the state of the lead vehicle has an associated
state uncertainty and does not take into account any information
from the current sample; (c) determining whether any of the object
points match the estimated state of the lead vehicle within the
state uncertainty; and (d) when at least one of the object points
matches the estimated state of the lead vehicle within the state
uncertainty, selecting the matching object point that best matches
the estimated state of the lead vehicle as a measured state of the
lead vehicle, and using the measured state of the lead vehicle in
the determination of a sequentially next estimate of a state of the
lead vehicle corresponding to a sequentially next sample; and
repeating steps (a)-(d) a multiplicities of times to thereby track
the lead vehicle.
23. A method as recited in claim 22 wherein the current state
estimate includes a plurality of state parameters, the state
parameters including a position parameter indicative of a position
of the lead vehicle relative to the trailing vehicle and a speed
parameter indicative of a velocity of the lead vehicle relative to
the trailing vehicle.
24. A method as recited in claim 22 further comprising at least
partially automatically controlling the trailing vehicle to
maintain a desired gap between the lead vehicle and the trailing
vehicle and wherein each selected object point has an associated
longitudinal distance from the distance measurement unit, and
wherein the associated longitudinal distance is treated by a gap
controller responsible for maintaining the desired gap as a current
measured longitudinal distance from the distance measurement unit
to the back of the lead vehicle.
25. A method as recited in claim 22 wherein: each sample indicates
a position of each of the object points; and each current estimate
of the state of the lead vehicle includes a current estimate of the
position of the lead vehicle and has an associated position
uncertainty; the selected matching object point must match the
estimated position of the lead vehicle within the position
uncertainty.
26. A method as recited in claim 25 wherein the current estimate of
the position of the lead vehicle estimates the current position of
a back of the lead vehicle.
27. A method as recited in claim 25 wherein the estimated position
of the lead vehicle is a relative position relative to the trailing
vehicle.
28. A method as recited in claim 25 wherein: each sample also
indicates a relative velocity of each of the object points; and
each current estimate of the state of the lead vehicle further
includes a current estimate of a relative velocity of the lead
vehicle and has an associated velocity uncertainty; the selected
matching object point must both (i) match the estimated position of
the lead vehicle within the position uncertainty, and (ii) match
the estimated velocity of the lead vehicle within the velocity
uncertainty.
29. A method as recited in claim 22 wherein when none of the object
points in a particular sample match the estimated state of the lead
vehicle within the state uncertainty, then the state uncertainty is
increased for the sequentially next estimate of the state of the
lead vehicle.
30. A method as recited in claim 29 wherein the estimate state
includes a plurality of state parameters, the state parameters
including a position parameter, a speed parameter and an
orientation parameter.
31. A method as recited in claim 22 further comprising:
periodically receiving global navigation satellite systems (GNSS)
position updates based at least in part on detected GNSS positions
of the lead and trailing vehicles; and each time a GNSS position
update is received, updating the estimated state of the lead
vehicle and the state uncertainty based on such GNSS position
update.
32. A method as recited in claim 22 further comprising:
periodically receiving vehicle speed updates based at least in part
on detected wheel speeds of the lead and trailing vehicles; and
each time a vehicle speed update is received, updating the
estimated state of the lead vehicle and the state uncertainty based
on such vehicle speed update.
33. A method as recited in claim 22 wherein steps (a)-(d) are
repeated at a sample rate of at least 10 Hertz.
34. A method as recited in claim 22 wherein Kalman filtering is
used to estimate the state of the lead vehicle and the associated
state uncertainty.
35. A method as recited in claim 22 wherein the estimated state of
the lead vehicle includes an estimated position of the back of the
lead vehicle and the selected matching object point is considered a
measurement of the relative position of the back of the lead
vehicle.
36. A method as recited in claim 22 wherein a controller on the
trailing vehicle maintains a profile of point clusters
representative of the lead vehicle and the selected matching point
corresponds to one of the point clusters.
37. A method as recited in claim 22 wherein the lead and trailing
vehicles are trucks involved in a platoon.
38. A method as recited in claim 22 wherein the distance
measurement unit is a radar unit.
39. A method of tracking a specific lead vehicle using a radar unit
mounted on a trailing vehicle, the method comprising: (a) obtaining
a current radar sample from the radar unit, the current radar
sample including a set of zero or more radar object points, each
radar object point indicating a relative a position of such radar
object point relative to the radar unit; (b) obtaining a current
estimate of a state of the lead vehicle corresponding to the
current radar sample, wherein the current estimate of the state of
the lead vehicle has an associated state uncertainty and includes a
current estimate of the position of a back of the lead vehicle
relative to the radar unit, wherein the current estimate of the
position of the back of lead vehicle has an associated position
uncertainty that is at least a part of the state uncertainty; (c)
determining whether any of the radar object points match the
estimated state of the lead vehicle within the state uncertainty,
wherein to match the estimated state of the lead vehicle within the
state uncertainty, a matching radar object point must match the
estimated position of the back of the lead vehicle within the
position uncertainty; and (d) when at least one of the radar object
points matches the estimated state of the lead vehicle within the
state uncertainty, selecting the matching radar object point that
best matches the estimated state of the lead vehicle as a measured
state of the lead vehicle, and using the measured state of the lead
vehicle in the determination of a sequentially next estimate of a
state of the lead vehicle corresponding to a sequentially next
radar sample; repeating steps (a)-(d) a multiplicities of times;
periodically receiving vehicle global navigation satellite systems
(GNSS) position updates based at least in part on detected GNSS
positions of the lead and trailing vehicles; each time a vehicle
GNSS position update is received, updating the estimated state of
the lead vehicle and the state uncertainty based on such vehicle
GNSS position update; periodically receiving vehicle speed updates
based at least in part on detected wheel speeds of the lead and
trailing vehicles; and each time a vehicle speed update is
received, updating the estimated state of the lead vehicle and the
state uncertainty based on such vehicle speed update; and at least
partially automatically controlling the trailing vehicle to
maintain a desired gap between the lead vehicle and the trailing
vehicle based at least in part on an aspect of the measured state
of the lead vehicle.
40. A method as recited in claim 39 wherein: each radar sample also
indicates a relative velocity of each of the radar object points;
and each current estimate of the state of the lead vehicle further
includes a current estimate of a relative velocity of the lead
vehicle and has an associated velocity uncertainty; the selected
matching radar object point must both (i) match the estimated
position of the lead vehicle within the position uncertainty, and
(ii) match the estimated velocity of the lead vehicle within the
velocity uncertainty.
41. A method as recited in claim 39 wherein: when none of the radar
object points in a particular radar sample match the estimated
position of the lead vehicle within the position uncertainty, then
the position uncertainty is increased for the sequentially next
estimate of the position of the lead vehicle; and when none of the
radar object points in a particular radar sample match an estimated
velocity of the lead vehicle within an velocity uncertainty, then
the velocity uncertainty is increased for the sequentially next
estimate of the position of the lead vehicle.
42. A method as recited in claim 39 wherein the estimated state of
the lead vehicle includes an estimated position of the back of the
lead vehicle and the selected matching radar object point is
considered a measurement of the relative position of the back of
the lead vehicle.
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] The present application is a continuation-in-part of PCT
Application No. PCT/US2016/060167, filed on Nov. 2, 2016, which
claims priority of U.S. Provisional Patent Application No.
62/249,898, filed on Nov. 2, 2015, both of which are incorporated
herein by reference in their entirety.
BACKGROUND
[0002] The present invention relates generally to systems and
methods for enabling vehicles to closely follow one another safely
using automatic or partially automatic control.
[0003] In recent years significant strides have been made in the
fields of autonomous and semi-autonomous vehicles. One segment of
vehicle automation relates to vehicular convoying systems that
enable vehicles to follow closely together in a safe, efficient and
convenient manner Following closely behind another vehicle has
significant fuel savings benefits, but is generally unsafe when
done manually by the driver. One type of vehicle convoying system
is sometimes referred to as vehicle platooning systems in which a
second, and potentially additional, vehicle(s) is/are autonomously
or semi-autonomously controlled to closely follow a lead vehicle in
a safe manner.
[0004] In vehicle platooning and convoying systems an understanding
of the distance between the vehicles is a very important control
parameter and multiple different independent mechanisms may be used
to determine the distance between vehicles. These may include radar
systems, transmitting absolute or relative position data between
vehicles (e.g., GPS or other GNSS data), LIDAR systems, cameras,
etc. A challenge that occurs when using radar in platooning type
applications is that the partner vehicle must be reliably
identified from a potentially ambiguous set of radar reflections
and tracked under constantly changing conditions. The present
application describes techniques for identifying and tracking
specific vehicles based on vehicle radar data that are well suited
for platooning, convoying and other autonomous or semi-autonomous
driving applications.
SUMMARY
[0005] A variety of methods, controllers and algorithms are
described for identifying the back of a particular vehicle (e.g., a
platoon partner) in a set of distance measurement scenes and/or for
tracking the back of such a vehicle. The described techniques can
be used in conjunction with a variety of different distance
measuring technologies including radar, LIDAR, sonar units or any
other time-of-flight distance measuring sensors, camera based
distance measuring units, and others.
[0006] In one aspect, a radar (or other distance measurement) scene
is received and first vehicle point candidates are identified at
least in part by comparing the relative position of the respective
detected objects that they represent, and in some circumstances the
relative velocity of such detected objects, to an estimated
position (and relative velocity) for the first vehicle. The first
vehicle point candidates are categorized based on their respective
distances of the detected objects that they represent from the
estimated position of the first vehicle. The categorization is
repeated for a multiplicity of samples so that the categorized
first vehicle point candidates include candidates from multiple
sequential samples. The back of the first vehicle is then
identified based at least in part of the categorization of the
first vehicle point candidates. The identified back of the first
vehicle or an effective vehicle length that is determined based at
least in part on the identified back of the first vehicle may then
be used in the control of the second vehicle.
[0007] In some embodiments, a bounding box is conceptually applied
around the estimated position of the first vehicle and measurement
system object points that are not located within the bounding box
are not considered first vehicle point candidates. In some
embodiments, the bounding box defines a region that exceeds a
maximum expected size of the first vehicle.
[0008] In some embodiments, the relative velocity of the vehicles
is estimated together with an associated speed uncertainty. In such
embodiments, object points within the set of detected object points
that are moving at a relative speed that is not within the speed
uncertainty of the estimated speed are not considered first vehicle
point candidates.
[0009] In some embodiments, categorizing the first vehicle point
candidates includes populating a histogram with the first vehicle
point candidates. The histogram including a plurality of bins, with
each bin representing a longitudinal distance range relative to the
estimated position of the first vehicle. In such embodiments, the
identification of the back of the first vehicle may be done after
the histogram contains at least a predetermined number of first
vehicle point candidates. In some embodiments, a clustering
algorithm (as for example a modified mean shift algorithm) is
applied to the first vehicle point candidates to identify one or
more clusters of first vehicle point candidates. In such
embodiments, the cluster located closest to the second vehicle that
includes at least a predetermined threshold percentage or number of
first vehicle radar point candidates may be selected to represent
the back of the first vehicle.
[0010] In some embodiments, Kalman filtering is used to estimate
the position of the first vehicle.
[0011] In another aspect, methods of tracking a specific lead
vehicle using a distance measuring unit mounted on a trailing
vehicle are described. In this embodiment, a current radar (or
other distance measurement) sample is obtained from a radar (or
other distance measurement) unit. The current distance measurement
sample includes a set of zero or more object points. In parallel, a
current estimate of a state of the lead vehicle corresponding to
the current sample is obtained. The current state estimate includes
one or more state parameters which may include (but is not limited
to), a position parameter (such as the current relative position of
the lead vehicle), a speed parameter (such as a current relative
velocity of the lead vehicle) and/or other position and/or
orientation related parameters.
[0012] The current estimate of the state of the lead vehicle has an
associated state uncertainty and does not take into account any
information from the current distance measurement sample. A
determination is made regarding whether any of the object points
match the estimated state of the lead vehicle within the state
uncertainty. If so, the matching object point that best matches the
estimated state of the lead vehicle is selected as a measured state
of the lead vehicle. That measured state of the lead vehicle is
then used in the determination of a sequentially next estimate of
the state of the lead vehicle corresponding to a sequentially next
sample. The foregoing steps are repeated a multiplicities of times
to thereby track the lead vehicle. The measured states of the lead
vehicle may be used in the control of one or both of the
vehicles--as for example in the context of vehicle platooning or
convoying systems, in the at least partially automatic control of
the trailing vehicle to maintain a desired gap between the lead
vehicle and the trailing vehicle.
[0013] In some embodiments, each sample indicates, for each of the
object points, a position of a detected object corresponding to
such object point (relative to the distance measuring unit). Each
current estimate of the state of the lead vehicle includes a
current estimate of the (relative) position of the lead vehicle and
has an associated position uncertainty. To be considered a valid
measurement, the selected matching object point must match the
estimated position of the lead vehicle within the position
uncertainty. In some implementations, the current estimate of the
position of the lead vehicle estimates the current position of a
back of the lead vehicle.
[0014] In some implementations, each sample indicates, for each of
the object points, a relative velocity of a detected object
corresponding to such object point (relative to the distance
measuring unit). Each current estimate of the state of the lead
vehicle includes a current estimate of the relative velocity of the
lead vehicle and has an associated velocity uncertainty. To be
considered a valid measurement, the selected matching object point
must match the estimated relative velocity of the lead vehicle
within the velocity uncertainty.
[0015] In some embodiments, when none of the radar object points in
a particular distance measurement sample match the estimated state
of the lead vehicle within the state uncertainty, then the state
uncertainty is increased for the sequentially next estimate of the
state of the lead vehicle.
[0016] In some embodiments, global navigation satellite systems
(GNSS) position updates are periodically received based at least in
part on detected GNSS positions of the lead and trailing vehicles.
Each time a vehicle GNSS position update is received, the estimated
state of the lead vehicle and the state uncertainty are updated
based on such position update.
[0017] In some embodiments vehicle speed updates are periodically
received based at least in part on detected wheel speeds of the
lead and trailing vehicles. Each time a vehicle speed update is
received, the estimated state of the lead vehicle and the state
uncertainty are updated based on such lead vehicle speed
update.
[0018] The described approaches are well suited for use in vehicle
platooning and/or vehicle convoying systems including
tractor-trailer truck platooning applications.
BRIEF DESCRIPTION OF THE DRAWINGS
[0019] The invention and the advantages thereof, may best be
understood by reference to the following description taken in
conjunction with the accompanying drawings in which:
[0020] FIG. 1 is a block diagram of a representative platooning
control architecture.
[0021] FIG. 2 is a flow chart illustrating a method of determining
the effective length of a platoon partner based on outputs of a
radar unit.
[0022] FIG. 3 is a diagrammatic illustration showing the nature of
a bounding box relative to a partner vehicle's expected
position.
[0023] FIG. 4A is a diagrammatic illustration showing exemplary
radar object points that might be identified by a radar unit
associated with a trailing truck that is following directly behind
a lead truck.
[0024] FIG. 4B is a diagrammatic illustration showing a
circumstance where the entire lead truck of FIG. 4A is not within
the radar unit's field of view.
[0025] FIG. 4C is a diagrammatic illustration showing a
circumstance where the bounding box associated with the lead truck
of FIG. 4A is not entirely within the radar unit's field of
view.
[0026] FIG. 4D is a diagrammatic illustration showing a
circumstance where the lead truck is in a different lane than the
trailing truck, but its entire bounding box is within the radar
unit's field of view.
[0027] FIG. 5A is a graph that illustrates the relative location
(longitudinally and laterally) of a first representative set of
partner vehicle radar point candidates that might be detected when
following a tractor-trailer rig.
[0028] FIG. 5B is a histogram representing the longitudinal
distances of the detected partner vehicle radar point candidates
illustrated in FIG. 5A.
[0029] FIG. 5C is a plot showing the mean shift centers of the
histogram points represented in FIG. 5B.
[0030] FIG. 5D is a graph that illustrates the relative location
(longitudinally and laterally) of a second (enlarged) set of
partner vehicle radar point candidates that might be detected when
following a tractor-trailer rig.
[0031] FIG. 5E is a histogram representing the longitudinal
distances of the detected partner vehicle radar point candidates
illustrated in FIG. 5D.
[0032] FIG. 5F is a plot showing the mean shift centers of the
histogram points represented in FIG. 5E.
[0033] FIG. 6 is a diagrammatic block diagram of a radar scene
processor suitable for use by a vehicle controller to interpret
received radar scenes.
[0034] FIG. 7 is a flow chart illustrating a method of determining
whether any particular radar scene reports the position of the back
of a partner vehicle and updating the estimator of FIG. 6.
[0035] FIG. 8 is a representation of a Kalman filter state array
and covariance matrix suitable for use in some embodiments.
[0036] In the drawings, like reference numerals are sometimes used
to designate like structural elements. It should also be
appreciated that the depictions in the figures are diagrammatic and
not to scale.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0037] The Applicant has proposed various vehicle platooning
systems in which a second, and potentially additional, vehicle(s)
is/are autonomously or semi-autonomously controlled to closely
follow a lead vehicle in a safe manner By way of example, U.S.
application Ser. Nos. 13/542,622, 13/542,627 and 14/292,583; U.S.
Provisional Application Nos. 61/505,076, 62/249,898, 62/343,819,
62/377,970 and; and PCT Application Nos. PCT/US2014/030770,
PCT/US2016/049143 and PCT/US2016/060167 describe various vehicle
platooning systems in which a trailing vehicle is at least
partially automatically controlled to closely follow a designated
lead vehicle. Each of these earlier applications is incorporated
herein by reference.
[0038] One of the goals of platooning is typically to maintain a
desired longitudinal distance between the platooning vehicles,
which is frequently referred to herein as the "desired gap". That
is, it is desirable for the trailing vehicle (e.g., a trailing
truck) to maintain a designated gap relative to a specific vehicle
(e.g., a lead truck). The vehicles involved in a platoon will
typically have sophisticated control systems suitable for
initiating a platoon, maintaining the gap under a wide variety of
different driving conditions, and gracefully dissolving the platoon
as appropriate.
[0039] The architecture and design of control systems suitable for
implementing vehicle platooning may vary widely. By way of example,
FIG. 1 diagrammatically illustrates a vehicle control architecture
that is suitable for use with platooning tractor-trailer trucks. In
the illustrated embodiment a platoon controller 110, receives
inputs from a number of sensors 130 on the tractor and/or one or
more trailers or other connected units, and a number of actuators
and actuator controllers 150 arranged to control operation of the
tractor's powertrain and other vehicle systems. An actuator
interface (not shown) may be provided to facilitate communications
between the platoon controller 110 and the actuator controllers
150. The platoon controller 110 also interacts with an
inter-vehicle communications controller 170 which orchestrates
communications with the platoon partner and a NOC communications
controller 180 that orchestrates communications with a network
operations center (NOC). The vehicle also preferably has selected
configuration files that include known information about the
vehicle.
[0040] Some of the functional components of the platoon controller
110 include gap regulator 112, mass estimator 114, radar tracker
116 and brake health monitor 118. In many applications, the platoon
controller 110 will include a variety of other components as
well.
[0041] Some of the sensors utilized by the platoon controller 110
may include GNSS (GPS) unit 131, wheel speed sensors 132, inertial
measurement devices 134, radar unit 137, LIDAR unit 138, cameras
139, accelerator pedal position sensor 141, steering wheel position
sensor 142, brake pedal position sensor 143, and various
accelerometers. Of course, not all of these sensors will be
available on all vehicles involved in a platoon and not all of
these sensors are required in any particular embodiment. A variety
of other sensor (now existing or later developed or commercially
deployed) may be additionally or alternatively be utilized by the
platoon controller in other embodiments. In the primary embodiments
described herein, GPS position data is used. However, GPS is just
one of the currently available global navigation satellite systems
(GNSS). Therefore, it should be appreciated that data from any
other GNSS system or from other suitable position sensing systems
may be used in place of, or in addition to the GPS system.
[0042] Many (but not all) of the described sensors, including wheel
speed sensors, 132, radar unit 137, accelerator pedal position
sensor 141, steering wheel position sensor 142, brake pedal
position sensor 143, and accelerometer 144 are relatively standard
equipment on newer trucks (tractors) used to pull semi-trailers.
However, others, such as the GNSS unit 131 and LIDAR unit 138 (if
used) are not currently standard equipment on such tractors or may
not be present on a particular vehicle and may be installed as
needed or desired to help support platooning.
[0043] Some of the vehicle actuators controllers 150 that the
platoon controller directs at least in part include torque request
controller 152 (which may be integrated in an ECU or power train
controller); transmission controller 154, brake controller 156 and
clutch controller 158.
[0044] The communications between vehicles may be directed over any
suitable channel and may be coordinated by inter-vehicle
communications controller 170. By way of example, the Dedicated
Short Range Communications (DSRC) protocol (e.g. the IEEE 802.11p
protocol), which is a two-way short to medium range wireless
communications technology that has been developed for vehicle to
vehicle communications, works well. Of course other communications
protocols and channels may be used in addition to or in place of a
DSRC link. For example, the inter vehicle communications may
additionally or alternatively be transmitted over a Citizen's Band
(CB) Radio channel, one or more General Mobile Radio Service (GMRS)
bands, and one or more Family Radio Service (1-RS) bands or any
other now existing or later developed communications channels using
any suitable communication protocol.
[0045] The specific information transmitted back and forth between
the vehicles may vary widely based on the needs of the platoon
controller. In various embodiments, the transmitted information may
include the current commands generated by the platoon controller
such as requested/commanded engine torque, requested/commanded
braking deceleration. They may also include steering commands, gear
commands, etc. when those aspects are controlled by platoon
controller. Corresponding information is received from the partner
vehicle, regardless of whether those commands are generated by a
platoon controller or other autonomous or semi-autonomous
controller on the partner vehicle (e.g., an adaptive cruise control
system (ACC) or a collision mitigation system (CMS)), or through
other or more traditional mechanisms--as for example, in response
to driver inputs (e.g., accelerator pedal position, brake position,
steering wheel position, etc.).
[0046] In many embodiments, much or all of the tractor sensor
information provided to platoon controller is also transmitted to
the platoon partner and corresponding information is received from
the platoon partner so that the platoon controllers on each vehicle
can develop an accurate model of what the partner vehicle is doing.
The same is true for any other relevant information that is
provided to the platoon controller, including any vehicle
configuration information that is relevant to the platoon
controller. It should be appreciated that the specific information
transmitted may vary widely based on the requirements of the
platoon controllers, the sensors and actuators available on the
respective vehicles, and the specific knowledge that each vehicle
may have about itself.
[0047] The information transmitted between vehicles may also
include information about intended future actions. For example, if
the lead vehicle knows it approaching a hill, it may expect to
increase its torque request (or decrease its torque request in the
context of a downhill) in the near future and that information can
be conveyed to a trailing vehicle for use as appropriate by the
platoon controller. Of course, there is a wide variety of other
information that can be used to foresee future torque or braking
requests and that information can be conveyed in a variety of
different forms. In some embodiments, the nature of the expected
events themselves can be indicated (e.g., a hill, or curve or exit
is approaching) together with the expected timing of such events.
In other embodiments, the intended future actions can be reported
in the context of expected control commands such as the expected
torques and/or other control parameters and the timing at which
such changes are expected. Of course, there are a wide variety of
different types of expected events that may be relevant to the
platoon control.
[0048] The communications between the vehicles and the NOC may be
transmitted over a variety of different networks, such as the
cellular network, various Wi-Fi networks, satellite communications
networks and/or any of a variety of other networks as appropriate.
The communications with the NOC may be coordinated by NOC
communications controller 180. The information transmitted to
and/or received from the NOC may vary widely based on the overall
system design. In some circumstances, the NOC may provide specific
control parameters such as a target gap tolerance. These control
parameters or constraints may be based on factors known at the NOC
such as speed limits, the nature of the road/terrain (e.g., hilly
vs. flat, winding vs. straight, etc.) weather conditions, traffic
or road conditions, etc. In other circumstances the NOC may provide
information such information to the platoon controller. The NOC may
also provide information about the partner vehicle including its
configuration information and any known relevant information about
its current operational state such as weight, trailer length,
etc.
Radar Tracking
[0049] The vehicles involved in a platoon will typically have one
or more radar systems that are used to detect nearby objects. Since
radar systems tend to be quite good at determining distances
between objects, separation distances reported by the radar unit(s)
are quite useful in controlling the gap between vehicles.
Therefore, once a platooning partner is identified, it is important
to locate that specific partner vehicle in the context of the radar
system output. That is, to determine which (if any) of a variety of
different objects that might be identified by the radar unit
correspond to the targeted partner.
[0050] Preliminarily, it should be appreciated that the platoon
partner will not always correlate to the closest vehicle detected
by the radar unit or to the vehicle that is directly in front of
the trailing truck. There are a wide variety of different scenarios
that can cause this to be the case. For example, when the platoon
is initially being set up, the partner may be out of sight of a
host vehicle's radar unit because it is too far away. As the
partner comes into sight of the radar unit, it becomes important to
identify and distinguish that partner from other objects in the
radar unit's field of view. The description below describes
techniques that are particularly well suited for identifying and
distinguishing a designated partner from other objects that may be
detected by a radar unit so that the radar unit can effectively
track the partner vehicle (sometimes referred to as "locking onto"
the partner).
[0051] Furthermore, during the course of driving, there will be
traffic in adjacent lanes that are traveling beside, passing or
being passed by the platoon and it is important for the radar unit
to be able to continue to differentiate the platoon partner from
passing vehicles so that the gap controller doesn't start trying to
maintain the gap from the wrong vehicle. In another example, a lead
truck may change lanes at which point it may not be directly in
front of the trailing vehicle, so again, it is important for that
the distance between the platoon partners reported by the radar
unit be associated with the platoon partner rather than merely the
closest vehicle or a vehicle that happens to be directly in front
of the trailing truck. There may also be times when the radar unit
may not be able to "see" the platooning partner. This could be
because an interloper has gotten between the platoon partners or
the lead vehicle has maneuvered out of view of the trailing
vehicle's radar unit, interference with the radar signals, etc.
[0052] For platoon control purposes, it is also important to
understand where the back of the vehicle is relative to the
vehicle's reported position. To elaborate, the position of the
partner vehicle is generally known from the GPS based location
information that is transmitted to the host vehicle. However, the
GPS system typically reports a location on the tractor, which could
for example, be the position of the antenna(s) that receive the GPS
signals. The detected GPS position may then be translated to the
position of a reference location on the vehicle that is a known
distance from the GPS antenna, with the position of that reference
location serving as the vehicle's reported GPS position. The
specific reference location chosen may vary based on control system
preferences. By way of example, in some tractor trailer truck
platooning embodiments, the reference location may be the center of
the rear axles of the tractor.
[0053] The difference between the reported GPS position and the
physical back of the vehicle can be significant to the platoon
control. Therefore, it is often important to know the distance
between the reported vehicle position and the actual back of the
vehicle. This is sometimes referred to herein as the "effective
vehicle length." The effective vehicle length is particularly
important in the context of a tractor trailer truck where the
reported GPS position is typically located somewhere on the cab
(tractor) and the distance from the reported GPS position to the
back of the trailer may be quite long. By way of example, trailer
lengths on the order of 12-18 meters are common in the U.S.
although they can be shorter or longer (indeed much longer in the
context of double or triple trailers). The distance from the
reported GPS position to the back of the vehicle must also account
for the longitudinal distance from the reported GPS position to the
front of the trailer and/or any extensions associate with the load.
It should be appreciated that in the trucking industry, the
effective vehicle length often will not be known since any
particular tractor may pull a variety of different trailers and the
attachment point between the tractor and trailer is adjustable on
the tractor.
Establishing a Radar Fix on a Platoon Partner
[0054] As will be apparent from the discussion above, a challenge
that occurs when using radar in platooning type applications is
that the partner vehicle must initially be found and identified in
the context of the radar system's output and thereafter reliably
tracked under constantly changing conditions. In application such
as the trucking industry, it is also desirable to determine the
effective length of at least the lead vehicle.
[0055] Commercially available radar units used in general road
vehicle driving automation systems typically output data that
indicates the presence of any object(s) detected within a
designated field together with the relative position and speed of
such object(s). Thus, during driving, such a radar unit may detect
the presence of a variety of objects within its operational field.
The detected objects may include any vehicle positioned directly in
front of the host vehicle, vehicles in adjacent lanes that may be
passing, being passed by or driving in parallel to the platoon,
stationary objects such as obstacles in the road, signs, trees, and
other objects to the side of the road, etc. Although many different
types of objects may be detected, the radar unit itself typically
doesn't know or convey the identity or nature of the detected
object. Rather it simply reports the relative position and motion
of any and all perceived objects within its operational field.
Therefore, to identify and track the partner vehicle in the context
of the radar unit output, it is helpful for the logic interpreting
the output of the radar unit to have and maintain a good
understanding of exactly where the partner vehicle is expected to
be relative to the radar unit's field of view regardless of whether
the partner vehicle is even in that field of view. This is possible
even when no explicit mechanism is provided for identifying the
partner because the platooning system preferably has multiple
independent mechanisms that can be used to help determine a
vehicle's position.
[0056] When a platoon partner is identified a communications link
is preferably established between the platooning vehicles. The
communications may be established over one or more wireless links
such as a Dedicated Short Range Communications (DSRC) link, a
cellular link, etc. Once communications are established between the
two vehicles, they begin transmitting data back and forth regarding
their respective selves, their current locations and operational
states. The processes used to identify potential platoon partners
and to establish the platoon and appropriate communication links
may vary widely. By way of example, a few representative techniques
are described in U.S. patent application Ser. Nos. 13/542,622 and
13/542,627 as well as PCT Patent Application Nos.
PCT/US2014/030770, PCT/US2016/049143 and PCT/US2016/060167
previously filed by Applicant, each of which is incorporated herein
by reference.
[0057] Once a platoon partner has been identified, the platoon
controller 110 requests the radar system control logic attempt to
find the partner vehicle. More specifically, the trailing vehicle's
radar tracker 116 needs to find and thereafter track the back of
the lead vehicle in the context of the radar unit's outputs so that
its data can be used in gap control. Referring next to FIG. 2, a
method particularly well suited for establishing a radar fix on a
platoon partner will be described. One aspect of establishing a
radar fix is to determine the length of the partner so the GPS
position information can be correlated to radar system outputs.
[0058] When the process initiates, radar tracker control logic
determines, receives or requests an estimate of the current
relative position of the partner vehicle and subscribes to or
regularly receives updates regarding the partner vehicle's relative
position as they become available as represented by step 203 of
FIG. 2. In addition to the relative locations, the estimated
information may optionally include various additional position
related information such as relative velocity of the vehicles, the
relative heading of the vehicles, etc.
[0059] In some embodiments, the radar tracker control logic is
configured to estimate the current relative position, velocity and
orientation (heading) of the partner vehicle based on a variety of
sensor inputs from both the host vehicle and the partner vehicle.
As mentioned above, the platoon partners are in communication with
one another and during platooning, they send extensive information
back and forth about themselves, including continually updated
information about their current location and operating states. By
way of example, some of the location related information that can
be helpful to interpreting radar unit data may include information
such as the partner vehicle's GPS position, wheel speed,
orientation/heading (direction that the vehicle is heading), yaw
rate (which indicates the vehicle's rate of turn), pitch, roll and
acceleration/deceleration (longitudinal and angular in any of the
forgoing directions). Operational related information may also
include a variety of other information of interest such the current
torque requests, brake inputs, gear, etc. Information about the
vehicles, may include information such as the make and model of the
vehicle, its length (if known), its equipment, estimated weight,
etc. Any of these and/or other available information can be used in
the position related estimates. By way of example, one particular
position estimator is described below with respect to FIGS. 6 and
7.
[0060] Although a particular estimator is described, it should be
appreciated that the estimated partner vehicle position related
information can come from any appropriate source and the estimation
does not need to be made by the radar tracker control logic itself.
Additionally, although it is preferred that position and
operational information be transmitted in both directions between
vehicles, that is not necessary as long as the host vehicle is able
to obtain the required information about the partner
vehicle(s).
[0061] The current location related information is updated very
frequently. Although the actual frequency of the updates can vary
widely based on the nature of the information being updated and the
nature of the communication link or vehicle system that provides
the information, update frequencies for items such as GPS position
and wheel speed received over a DSRC link at frequencies on the
order of 10 to 500 Hz, as for example 50 Hz work well although
slower and much faster update frequencies may be used as
appropriate in other embodiments. Furthermore, although regular
updates of the location related information are desirable, there is
no need that they be received synchronously or at consistent
intervals.
[0062] It should be appreciated that when the radar system begins
trying to locate the partner vehicle, the partner vehicles may or
may not be within the radar unit's field of view. However both the
host vehicle's position and the partner vehicle's position are
generally known based at least on the received GPS data so it is
easy to estimate their separation with reasonable certainty. It
should also be appreciated that although GPS location signals tend
to be pretty good, the reported locations may be off by some amount
and thus it is better to treat any reported GPS position as an
estimate with some appropriate amount of uncertainty rather than
treating the reported position as infallible information. More
details regarding some specific algorithms that are suitable for
estimating the partner vehicle position will be described in more
detail below. Experience has shown that GPS position readings from
commercially available GPS sensors used in vehicle automation
applications tend to be accurate within about 2-3 meters in
practical road conditions when there is a direct line of sight to
at least 4 GPS satellites. However, it should be appreciated that
some GPS sensors are regularly more precise and no GPS sensors are
guaranteed to always be that accurate due to variables such as
interference, operations is regions where there is not line of
sight visibility to the required number of operational GPS
satellites, etc.
[0063] Once the partner vehicle's relative position estimate is
known, a bounding box is applied around the estimated relative
position of the partner (step 206 of FIG. 2). The purpose of the
bounding box is to define a region that the partner vehicle is
"expected" to be found in. The logic will thereafter look for radar
detected objects located within that bounding box in an effort to
identify objects that may correlate to the partner vehicle. The
concept of a bounding box is helpful for several reasons. Initially
it should be appreciated that the GPS unit will typically report
the location of its antenna, which in the context of a
tractor-trailer truck is usually on the cab. This detected position
is then typically translated to a predefined reference location on
the tractor and that translated position is used as the reported
GPS position. Thus, the reported GPS position for a tractor-trailer
will be well in front of the back of the trailer which is (a) the
point that is of primary interest to the gap control purposes, and
(b) is typically the most prominent feature identified by the radar
unit from a trailing platoon partner. Furthermore, the distance
between the reported GPS position and the back of the trailer will
not be known in many circumstances. One reason for the uncertainty
is that a particular tractor (cab) may be used to pull a variety of
different trailers (or other loads) which potentially have
different lengths. Therefore the effective length of the
tractor-trailer combination may vary from trip to trip and from a
control standpoint it is generally undesirable to count on the
driver to manually input the effective length of the
tractor-trailer combination each trip. To a lesser extent the
reported GPS positions of both platoon partners are subject to a
degree of uncertainty.
[0064] The actual size and geometry of the bounding box used may
vary but it is desirable that the region be large enough to
encompass the entire range of vehicle lengths and widths that are
possible plus a buffer to account of uncertainty in the estimated
GPS position. Thus, for trucking applications, it is desirable that
the longitudinal length of the bounding box be longer than any
tractor-trailer combination that might be expected to be
encountered. For example, U.S. commercial trucking applications
involving normal tractor trailer combinations typically don't
significantly exceed a combined length of 23 meters. In such
applications, bounding boxes on the order of 32 meters long and
3-4.5 meters, as for example 3.8 meters wide have been found to
work well. In regions that allow longer trailers or the use of
double or triple trailers, the tractor-trailer combinations may be
longer and therefore longer bounding boxes may be appropriate. If
the actual length of the platoon partner is known, the size of the
bounding box can be adjusted accordingly to more accurately reflect
the expected offset between the GPS position and the back of the
trailer--which correlates to the effective vehicle length. However,
even when it is believed that the effective length and width of the
platoon partner is "known," it is still desirable to utilize a
bounding box greater in size than the reported length and width to
accommodate uncertainty in the GPS estimates and the possibility
that the load may include a feature that extends beyond the
vehicle's reported length.
[0065] It should be appreciated though that there is no need for
the bounding box to be rectilinear in nature, rather, the bounding
box may encompass any desired geometric shape and/or may include
dimensions other than longitudinal length and lateral width--as for
example relative velocity. Thus, the bounding box may be defined in
any desired manner.
[0066] A representative bounding box 255 applied around a lead
truck 251 in a platoon of two trucks is diagrammatically
illustrated in FIG. 3. In the illustrated embodiment, each truck
has a GPS unit 258 located on its tractor (cab) and a radar unit
260 located at the front of the cab. It can be seen that the
bounding box exceeds the length and width of the lead truck
251.
[0067] In some embodiments, the bounding box may be defined more
complexly. For example, in one particular embodiment, the scaled
squares of the lateral offset (Y.sub.off) and the relative velocity
(V) of the vehicles may be compared to a threshold (Th). A radar
point would then be rejected if the sum of these squares exceeds
the designated threshold (Th), even if the radar point is within
the longitudinal range of the bounding box. Such a test may be
represented mathematically as shown below:
If kY.sub.off.sup.2+V.sup.2.gtoreq.Th, then the object is
rejected
In such an approach, the bounding box has the effective appearance
of a tube with in a state space map with velocity being the third
axis. The logic of such an approach is that if both the measured
lateral offset and the measured velocity of a detected object are
relatively lower probability matches, then the detected point is
less likely to be a match (and therefore more appropriate to
disregard for the purposes of identifying the back of a partner
vehicle) than if one of those parameters is off but the other very
nearly matches the expected value. Although only a couple specific
bounding box definition approaches have been described, it should
be apparent that a wide variety of other bounding box definitions
may be used as appropriate in other implementations. Additionally,
the bounding box definition may be arranged to change over time.
For example, one or more selected dimensions of the bounding box
may be reduced as the algorithm begins to develop a better
understanding of what radar object sample points are more likely to
correspond to the partner vehicle or the back of the partner
vehicle.
[0068] Once the bounding box has been established, the logic
determines whether the entire bounding box is within the other
vehicle's radar unit's field of view 263 (step 209). If not, the
logic waits for the entire bounding box to come within the radar
unit's field of view thereby effectively ignoring the radar system
outputs for the purpose of identifying the partner vehicle
(although of course the radar system outputs can be used for other
purposes such as collision avoidance if desired). There are a
variety of reasons why the partner vehicle may not be within or
fully within the radar units field of view at any particular time.
Initially, it should be appreciated that although the radar unit(s)
used to support platooning may be placed at a variety of different
locations on the vehicles, they often have a relatively narrow
field of view. For example, one common approach is to place a
forward facing radar unit having a relatively narrow fixed beam in
the vicinity of the middle of the front bumper to detect objects in
front of the vehicle. Such an arrangement is illustrated in FIG. 3.
In that figure, the field of view 263 of radar unit 260 located on
the trailing truck 252 is also shown.
[0069] When a forward facing radar unit is used, it will be unable
to see any vehicle behind or to the side of its host vehicle. Even
when the partner vehicle is ahead of the radar unit host, it may be
out of the field of view if it is too far ahead of the host or is
around a corner--as may be the case when a platoon partner is first
identified. In some cases a platoon partner can be partially in the
radar unit's field of view. A common example of this is when the
partner vehicle in an adjacent lane and not far enough ahead for
the back of its trailer to be seen by a narrow beamed forward
facing radar unit. It should be appreciated that it is undesirable
to utilize radar samples if the back of the bounding box is not
within the radar unit's field of view, since there is a risk that
the furthest back portion of the partner vehicle that is detected
by the radar unit is not actually the back of the vehicle.
[0070] FIGS. 4A-4D illustrate a few (of the many) potential
relative positioning of two trucks that are in the process of
establishing a platoon. In FIG. 4A, the lead truck 251 is directly
ahead of the trailing truck 252 and its bounding box 255 is fully
within the field of view 263 of trailing truck radar unit 260. In
contrast, in FIG. 4B, the lead truck 251 is in a lane adjacent the
trailing truck 252 and some, but not all of the lead truck 251
itself (and thus not all of bounding box 255) is within the field
of view 263 of trailing truck radar unit 260. In FIG. 4C, the lead
truck 251 is in a lane adjacent to the trailing truck 252 and all
of the lead truck 251 itself, but not the entire bounding box 255,
is within the field of view 263 of trailing truck radar unit 260.
In FIG. 4D, the lead truck 251 is again in a lane adjacent the
trailing truck 252 but differs from FIGS. 4B and 4C in that the
entire bounding box 255 associated with lead truck 251 is within
the field of view 263 of trailing truck radar unit 260. In
circumstances where the entire bounding box is not located within
the radar unit's field of view (e.g., a scenario such as shown in
FIG. 4B or 4C or when the lead vehicle is otherwise out of view),
the partner vehicle identification logic waits at step 209 for the
entire bounding box to come within the radar unit's field.
[0071] When the entire bounding box is within the radar unit's
field of view (e.g. a scenario such as illustrated in FIG. 4A or
FIG. 4D), the radar system controller logic obtains a next radar
sample (step 212) and a current estimate of the partner vehicle's
position and velocity relative to itself (step 215). Commercially
available short range radar units utilized in road vehicle
applications are typically configured to output their sensed scene
at a relatively rapid sample rate. Each scene typically identifies
a set of zero or more objects that have been detected as well as
the velocity of such objects relative to the radar unit itself.
[0072] The nature of radar systems is that the transmitted radio
waves can be reflected by most anything in their path including
both any intended target(s) and potentially a wide variety of
different items. Therefore, when trying to establish a platoon, it
is important to identify the reflected signal(s) that represent the
desired partner and to be able to distinguish that partner from the
noise reflected from other objects. By way of example, when driving
along a road, the radar unit may receive reflections from multiple
different vehicles including any vehicle that is immediately ahead,
passing vehicles going in the same or opposite direction objects to
the side of the road such as highway or street signs, trees or
other objects along the side of the road, etc.
[0073] When a sensed scene is received, the radar system control
logic determines whether any of the identified objects are partner
vehicle radar point candidates as represented by step 218.
Representative objects that might be detected by the radar unit 260
are marked with X's in FIGS. 4A-4D. To qualify as a partner vehicle
radar point candidate, an object detected in the scene must be
located within the bounding box in terms of both position and
speed. Radar objects located outside of the bounding box are
preferably rejected because there is a relatively higher
probability that they do not correspond to the partner vehicle. For
example, they could correspond to vehicles in adjacent lanes 272,
273, an interloper located between the platoon partners (not
shown), objects on the side of the road 274, etc. Objects that do
not closely match the expected relative speed of the partner
vehicle are also preferably rejected even if they match the
expected position aspects of the bounding box longitudinally and
laterally because again, it is less likely that they correspond to
the platoon partner. For example, a stationary object such as a
feature to the side of the road (e.g. a road sign, tree or
stationary vehicle), debris in the road, or a detected feature in
the road itself (e.g. a pothole, etc.), will appear to be
approaching the radar unit at the speed that the host vehicle is
traveling at. It is noted that many commercially available radar
units will automatically filter out, and therefore don't report,
stationary objects. When such a radar unit is used, the stationary
objects would not even be identified as part of the radar
scene.
[0074] Some of the reported radar objects may be traveling in the
same direction as the host vehicle but are moving at a relative
velocity that is different than the expected partner velocity.
There is a relatively high probability that such radar objects do
not correspond to the partner vehicle and therefore these types of
radar points are also preferably discarded.
[0075] Any detected radar objects that appear to match the expected
location and speed of the partner within the context of the defined
bounding box are considered partner vehicle radar point candidates
and are categorized with respect to how far they are longitudinally
(along the longitudinal axis of the partner) from the estimated
location of the partner (e.g., the partner's GPS position). In some
embodiments, a histogram is utilized for to this categorization.
The number of bins in the histogram may vary. For computational
ease, 512 bins divided evenly over the length of the bounding box
has been found to work well, although more or less bins can be used
as appropriate for any particular application. In implementations
that use a bounding box of approximately 32 meters, with 512 bins,
each bin corresponds to approximately 6 cm (2-3 inches). If greater
resolution is desired, then more bins can be used.
[0076] It has been observed that it is common for the short range
radar units utilized in road vehicle applications to identify
multiple different "objects" that may be actually part of the same
vehicle as represented by radar points 276-279 in FIGS. 4A-4D. This
is particularly common in trucks and indeed it is common for the
radar signature of a tractor-trailer truck to appear as more than
one object. For example, the back of the trailer, an underride
guard, and/or other features of the trailer or load located near
the back of the trailer may appear in the radar output as one or
multiple distinct objects (e.g., points 276, 277). Additionally,
objects located further up the trailer and/or objects in the
vicinity of the cab may be separately identified (e.g. points 278,
279). For example when the radar is mounted relatively low on the
host vehicle it may detect reflections from the transmission or
other items along the truck's undercarriage or other features of
the tractor-trailer such as the trailer's landing gear or the back
of the tractor and identify those items as separate detected
"objects." Therefore, it is possible (indeed it is relatively
common) that any particular sample may identify more than one
object that meets the criteria of a partner vehicle radar point
candidates. In such circumstances multiple candidates associated
with a particular radar sample will be added to the histogram.
[0077] After the histogram has been populated with any partner
vehicle radar point candidates identified in the sample, a
determination is made regarding whether sufficient samples have
been obtained to analyze the radar data to identify the partner
vehicle in step 224. If not, the logic returns to step 212 where
the next sample is obtained and the process repeats until
sufficient samples have been obtained to facilitate analysis. If
the bounding box moves partially out of the field of view of the
radar unit at any point (as represented by the "no" branch from
decision block 225), then the logic returns to step 209 where it
waits for the bounding box to come back into full view before
taking additional samples.
[0078] As discussed above, commercially available short range radar
units utilized in road vehicle applications are typically
configured to output their sensed scene at a relatively rapid
sample rate. By way of example, sample rates on the order of 20 to
25 hertz are common, although either higher or lower sample
frequencies may be used. Therefore, the histogram will populate
fairly quickly when the partner vehicle is within the radar unit's
field of view and the histogram will provide a rather good
indication of the radar signature of the partner.
[0079] FIG. 5A is a plot showing a set of 98 detected partner
vehicle radar point candidates transposed into a reference frame
based on the expected location of the front truck. The x-axis of
the plot shows the longitudinal distance from the expected position
of the front of the leading truck to the detected point. The y-axis
shows the lateral offset of the detected point relative to the
center axis of the leading truck. It can be seen that although
there is noticeable variation in the locations of the detected
points, in the illustrated sample set, the points tend to be
clustered into a couple of regions. FIG. 5B is a histogram that
shows the longitudinal distance to each of the detected partner
vehicle radar point candidates in the plot of FIG. 5A. It can be
seen that when only the longitudinal distance is considered, the
clustering tends to be even more pronounced.
[0080] The large cluster 290 located furthest back in the histogram
typically corresponds to the back of the vehicle and is often
(although not always) the largest cluster. Cluster 292 located
further forward typically correspond to other features of the
partner truck. Experience has shown that radar reflections from the
forward features tend to be weaker and more sporadically identified
as a discrete object by the radar unit, which translates to a
smaller cluster in the histogram.
[0081] If sufficient samples have been obtained to support
analysis, the logic follows the yes branch from decision block 224
and flows to step 227 where a clustering algorithm is applied to
the histogram data. The trigger point for when processing may start
can vary widely based on the needs of any particular system. In
general, it is desirable for the histogram to contain enough data
points so that the partner vehicle can be accurately identified. In
some specific implementations, the histogram must include data from
a first threshold worth of samples (e.g., samples corresponding to
at least 3 seconds worth of data or 60 samples) and include at
least a second threshold worth of partner vehicle radar point
candidates (e.g., at least 60 partner vehicle radar points). The
thresholds used may vary based on the needs of a particular
implementation. By way of example, samples corresponding to at
least 1-5 seconds worth of data or thresholds in the range of 40 to
500 points may be used in some implementations. In one specific
example, samples corresponding to at least 3 seconds worth of data
or 60 samples and 60 partner vehicle radar points are used as
thresholds.
[0082] The dataset illustrated in FIGS. 5A and 5B is representative
of a dataset that might be available at the time that an attempt is
initially made to identify the back of the partner vehicle--that
is, the first time that the "yes" branch from step 224 is
followed.
[0083] In general, the clustering algorithm bunches data points
that are highly likely to represent the same point. A variety of
conventional clustering algorithms can be used for this purpose. By
way of example, modified mean shift algorithms work well. FIG. 5C
is a plot showing the mean shift centers of the histogram points
represented in FIG. 5B, with the heights of the centers being
indicative of the number of points associated with that center. The
two clusters 290 and 292 stand out even more dramatically in this
representation.
[0084] The mean shift data is then analyzed to determine whether
one of the clusters meets predefined back of partner vehicle
criteria in step 230. If so, that cluster is identified as
corresponding to the back of the vehicle. (Step 233). Since each
cluster corresponds to a designated distance between the partner's
reported GPS position and the back of the vehicle, the effective
length of the vehicle is defined by the cluster. As noted above,
the phrase "effective vehicle length" as used herein corresponds to
the distance between the reported GPS position and the back of the
vehicle--which is an important distance to know for control
purposes. It should be appreciated that this is typically different
than the actual length of the vehicle because the reported
reference position may not be located at the front of the
vehicle.
[0085] In some implementations the cluster located closest to the
back of bounding box that has over a threshold percentage of the
total number of radar points in the histogram is identified as back
of the platoon partner vehicle. In some implementations a further
constraint is used that requires that the cluster location not move
by more than a certain threshold on the last sample. By way of
example, maximum movement thresholds on the order of 1 mm have been
found to work well in some applications. This approach has been
found to very reliably identify the radar point that corresponds to
the back of a truck even when the radar unit controller has no
predetermined knowledge of the length of the vehicle and regardless
of the presence of other traffic. However, it should be appreciated
that the threshold percentage or other characteristics of the
histogram used to identify the back of the vehicle may vary based
on application. In the embodiment illustrated in FIGS. 5A-5C,
cluster 290 is designated as the back of the lead truck.
[0086] It is particularly noteworthy that even though other traffic
moving in parallel with the platoon may be detected by the radar,
the described approach very reliably filters those radar points by
effectively applying a number of different types of filters. Radar
points that report features that are not where the platoon partner
is expected to be are filtered because they are not within the
bounding box. Radar points that are not traveling at close to the
expected relative speed are filtered regardless of where they are
found. The back of vehicle criteria used on the clustered histogram
data effectively filters any other vehicles traveling within the
footprint of the bounding box at very near the same speed as the
platoon partner because the bins are small enough that it is highly
unlikely that such an interloper can maintain a constant enough gap
to fool the algorithm into thinking that the interloper is part of
the target (e.g., even if the interloper is traveling at nearly the
same speed as the partner vehicle, if it is located within the
bounding box, it's position relative to the partner vehicle's
position is likely to vary enough to cause the back of partner
vehicle test to fail. The back of vehicle criteria also filters out
more random objects reported by the radar unit.
[0087] The effective vehicle length indicated by the selected mean
shift cluster may be reported to the gap controller and any other
controller concerned with the length of the partner. In most
circumstances, the distance between the GPS reference location and
the front of the host vehicle is known and therefore the effective
vehicle length determined by the radar unit can readily be used in
association with known information about the truck to positively
indicate the front and back of the truck as represented by step
236.
[0088] In some circumstances none of the mean shift clusters will
meet the back of partner vehicle criteria. In most cases this
suggests that there is a risk that the partner vehicle is not being
accurately tracked. In such cases (as illustrated by the no branch
from decision 230) the process continues to run collecting radar
points from additional samples until the criteria is met indicating
that the partner vehicle has confidently been identified. In some
embodiments, radar points may optionally be discarded after they
become too old or the process restarted if the system has trouble
identifying the back of the partner vehicle or for other reasons,
such as the vehicles coming to a stop.
[0089] In some embodiments, the back of the partner identification
process continues to run or is periodically rerun even after the
vehicle length has been determined. There are several advantages to
continuing to populate the histogram. Often the initial length
determination is made while the platoon partners are relatively far
apart (e.g., over 100 feet). Once the back of the partner vehicle
has been reliably identified, the gap controller may tighten the
gap thereby drawing the vehicles closer together. When the vehicles
are closer together, the radar reading are often more precise than
they are when the vehicles are 100+ feet apart. Additionally,
remembering that in some circumstances the GPS measurements may be
relatively far off for gap control purposes, more measurement give
a better statistical indication of the relative position of the
vehicle. By continuing to run the back of partner identification
process, those better measurements can be used to more accurately
determine the effective length of the partner vehicle, which is
highly desirable for control purposes.
[0090] FIG. 5D is a plot showing a set of 1700 detected partner
vehicle radar point candidates on the same graph as shown in FIG.
5A. The 1700 sample points include the 98 points illustrated in
FIGS. 5A-5C and were obtained by continuing to run the same radar
point classification algorithm. FIGS. 5E and 5F show the histogram
and mean shift centers respectively for the larger data set. Thus,
FIG. 5E corresponds to FIG. 5B, and FIG. 5F corresponds to FIG. 5C.
It can be seen that the larger dataset appears to have identified a
small cluster 293 located near the front of the lead vehicle and
has effectively filtered out some smaller clusters identified in
the smaller data set.
[0091] Continuing to run the back of partner identification process
has other potential uses as well. For example, some trucks have the
ability to draw the trailer closer to the cab when the truck is
operating on the highway. Thus, although it is relatively rare,
there are situations in which the effective length of the truck can
vary over the course of a platoon. Such changes can automatically
be detected by rerunning or continuing to run the back of the
partner identification process.
[0092] Over time, the histogram and/or mean shift clusters also
provide a very good indication of the radar signature of the
partner vehicle. This known signature of the partner vehicle can be
used in a number of different ways as an independent mechanism for
verifying that the proper vehicle is being tracked. For example, in
scenarios where GPS data becomes unavailable or communications
between the vehicles are disrupted for a period of time, the
histogram can be used as a check to verify that the correct vehicle
is being tracked by the radar unit. In circumstances where the back
of the lead truck is not within the view of the trailing vehicle's
radar, but other portions of the trailer and tractor are within the
radar's view, the portion of the truck that can be seen can be
compared to the histogram signature to determine the relative
positioning of the trucks, which can be used as a measurement for
gap control or as part of autonomous or semi-autonomous control of
the trailing vehicle.
[0093] In another example, in circumstances when radar contact is
lost, a new histogram can be started at an appropriate time and a
new histogram can be compared to a stored histogram indicative of
the platoon partner. When there is a match, that match can be good
independent evidence that radar contact with the platoon partner
has been reestablished. Similarly, newly created histograms can be
compared to stored histograms representing the platoon partner at
various times during platooning as a way of independently verifying
that the platoon partner is still being tracked. This can be a good
safety check to verify that the radar unit has not inadvertently
switched and locked onto a vehicle that is traveling in parallel
next to the platoon partner. The histograms can also be saved as a
radar signature of the partner vehicle and shared with other trucks
that may later seek to platoon with that vehicle--which can be
useful in the initial identification process.
Estimating Position of Platoon Partners
[0094] In the context of platooning, it is helpful to maintain
accurate models of the expected relative positions, speeds and
orientations of each of the vehicles in the platoon as such
information is very helpful in the accurate control of the gap
between platoon partners. Such models preferably utilize inputs
from multiple different sensing systems and include at least some
redundant information from different systems when practical. The
provision of redundant information from different systems is
helpful as a double check as to the integrity of received data and
also provides backup mechanisms for the inevitable times when a
system is unable to convey accurate information.
[0095] By way of example, the gap between vehicles can be
determined using a number of different techniques. One general
approach is to use the distance to the platoon partner detected by
the radar system. Although radar tends to very accurately measure
the distance between vehicles, it is important to ensure that the
distance being reported is actually the distance to the platoon
partner rather than some other vehicle or feature. There are also
times when the partner vehicle is not within the radar's field of
view or the radar or the radar unit is not operating as desired for
a brief period. An independent way of determining the distance
between the platoon partners is to utilize their respective GPS
data. Specifically, the distance between the vehicles should be the
difference between the vehicle's respective GPS positions, minus
the effective length of the lead vehicle and the offset distance
between the front of the trailing vehicle and its GPS receiver.
Limitations of using the GPS data include the fact that the GPS
data will not always be available due to factors such as the GPS
receivers not having a clear view of sufficient GPS satellites to
be able to determine a location or the communication link between
vehicles being down for a period of time. The GPS data is also
fundamentally limited by the fact that the accuracy of the GPS
data, which while good, is often less precise than desired for gap
control. Other systems for measuring distances between the platoon
partners have their own advantages and limitations.
[0096] When the current gap between the vehicles is known, the gap
expected at a time in the immediate future can be estimated based
on factors such as the current positions, the relative velocities
and yaw rates of the vehicles. The respective velocities of the
vehicles may also be measured, determined, estimated and/or
predicted in a variety of different manners. For example, wheel
speed sensors can be used to relatively accurately indicate the
current speeds of the respective vehicles. Knowledge of the
vehicle's orientation can be used in conjunction with the knowledge
of the vehicle's speed to determine its velocity. The radar unit
can be used to measure the relative speeds of the platoon partners.
Knowledge of other factors such as torque request, vehicle weight,
engine characteristics and road grade can be used to predict
vehicle speeds in the future.
[0097] In the context of the radar system control, knowing where
the leading vehicle is expected to be relative to the radar unit on
a trailing vehicle can be quite helpful in determining whether one
or more objects detected by the radar unit correspond to the back
of the lead vehicle. Therefore, in some embodiments, the radar
system controller (or another controller whose determinations can
be utilized by the radar system controller) includes a position
estimator that maintains an estimate of the current position,
orientation and relative speed of the partner vehicle relative to
the radar unit. One suitable radar scene processor 600 that
includes a position/state estimator 612 is illustrated in FIG.
6.
[0098] In the embodiment illustrated in FIG. 6, radar scene
processor 600 includes gap monitor 610 and a partner identifier
620. The gap monitor 610 is configured to track the position of the
back of the partner vehicle based on radar measurements (after the
back of the partner vehicle has been identified) and to report
radar position and speed measurements corresponding to the back of
the partner vehicle to the gap controller and/or any other
component interested in such measurements made by the radar unit.
One particular implementation of the gap monitoring algorithm will
be described below with reference to the flow chart of FIG. 7.
[0099] In the illustrated embodiment, the gap monitor 610 includes
a position/state estimator 612 having a Kalman filter 615 that is
used to determine both the most recent estimate of the position of
the partner vehicle relative to the host vehicle and to predict the
expected position of the partner vehicle at the time the next radar
sample will be taken. As described in more detail with respect to
FIG. 7, in the illustrated embodiment, the position/state estimator
612 utilizes both the detected radar scenes and other available
vehicle state information such as the respective GPS positions,
wheel speeds, and inertial measurements of the host and partner
vehicles in the estimate of the expected state (e.g. position,
velocity etc.) of the leading vehicle. These state estimates can
then be used to help interpret the received radar scene. That is,
having a reasonable estimate of where the partner vehicle is likely
to be in the context of a radar scene helps the gap monitor 600
properly identify the radar return object that corresponds to the
back of the partner vehicle out of a radar scene that may include a
set of detected objects. This helps ensure that the proper detected
point is used in the gap control. It is also helpful in identifying
situations in which the tracker does not have good confidence
regarding which (if any) of the objects detected by the radar in a
particular scene sample accurately represent the position of the
back of the partner vehicle so that such a sample can be
discounted, ignored or otherwise properly handled in the context of
the gap control algorithm. One particular Kalman filter design that
is well suited for use in the position/state estimator 612 is
described below with respect to FIG. 8.
[0100] The partner identifier 620 includes its own position/state
estimator 622, a histogram 624, a clustering algorithm 625 which
produces mean shift clusters 626 and partner length estimator 627.
The partner identifier 620 executes an algorithm such as the
algorithm discussed above with respect to FIG. 2 to identify the
back of the partner vehicle. As part of that process, histogram 624
is populated. The histogram is diagrammatically shown as being part
of the partner identifier 620, but it should be appreciated that
the histogram is merely a data structure that can be physically
located at any appropriate location and may be made available to a
variety of other processes and controllers within, or external to,
the radar tracker 620. The partner length estimator 624 is
configured to determine the length of the partner vehicle
(including its front and back relative to its GPS reference
position) based on the histogram and other available
information.
[0101] The position/state estimator 622 in the partner identifier
620 functions similarly to the position/state estimator 612
describe above and may also include a Kalman filter 623. A
significant difference between position state estimator 622 used
for partner identification and position/state estimator 612 is that
what radar point corresponds to the back of the partner truck is
not known during identification and therefore the radar unit
samples cannot be used as part of the position/state estimates.
[0102] The position/state estimation, partner detection, partner
length estimating and gap monitoring algorithms may be executed on
a radar tracking processor dedicated to radar tracking alone, or
they may be implemented on a processor that performs other gap or
platoon management tasks as well. The respective algorithms may be
implemented as distinct computing processes or they may be
integrated in various manners with each other and/or other
functionality in various computing processes. In other embodiments,
discrete or programmable logic may be used to implement the
described functionality. It should be apparent that a wide variety
of different models can be used to track the position of the back
of the partner vehicle relative to the radar unit and to estimate
future positions. Two particular position/state estimators are
diagrammatically illustrated as part of FIG. 6 and a method that
can be used to estimate the current position at any given radar
sample time is illustrated in the flow chart of FIG. 7.
[0103] Referring next to FIG. 7, a method of tracking a partner
vehicle and estimating its future position based in part on
information received from the radar unit will be described. In the
illustrated embodiment, the trailing vehicle is tracking the
position of the back of a lead vehicle, although an analogous
process can be used by the lead vehicle to track a following
vehicle or for parallel vehicles to track one another. The
described method presupposes that we have a reasonable estimate of
the location of the back of the partner vehicle--which can
initially be determined using the method described above with
respect to FIG. 2 or in any other suitable manner. For example,
when the effective length of the front vehicle is known, the
initial estimate for the relative position of the back of the lead
vehicle can be estimated based on GPS position data.
[0104] Each time a new radar scene is received (step 502) a
determination is made regarding whether any of the radar object
points (targets) matches the expected position and relative
velocity of the back of the partner vehicle (step 504). This is
preferably a probabilistic determination in which it is concluded
that that there is a high probability that the "matching" target
indeed represents the back of the partner vehicle. One way to
determine whether a matching target is to quantify an uncertainty
factor in association with the estimated position. If a radar
target point is within the range of the uncertainty factor of the
expected position, then it can be considered a match. As will be
described in more detail below in some implementations Kalman
filtering is used to estimate the position of the back of the
partner vehicle and to quantify the uncertainty. Kalman filtering
is particularly appropriate because it inherently adjusts the
uncertainty level based on the perceived accuracy of the
measurements.
[0105] If more than one of the reported radar target points match
the estimated position within the range defined by the uncertainty
factor (sometimes referred to as a ball of uncertainty), then the
closest radar object point identified in the radar scene is treated
as the "matching" target. In the context of this determination, the
"closest" match may be selected based on a combination of metrics
including longitudinal position, lateral position, relative speeds,
etc.
[0106] If a match is found, the radar tracker transmits the
distance to the matched object and relative speed of the matched
object to the gap controller 112 as the current gap to and relative
speed of, the back of partner vehicle (step 506). In some
embodiments, the only information transmitted is the longitudinal
distance to the back of the trailer and its relative speed. This is
because while currently available radar units are generally quite
good at measuring distance and relative speed, they are not as good
at precisely measuring lateral velocities or providing precise
lateral position information regarding identified objects. However,
if the radar unit used can accurately measure other useful
attributes of the target such as lateral velocities, acceleration,
etc.,--that information may optionally be transmitted as well.
[0107] When a match is found, the best matched target is used to
update the radar tracking position and speed estimate for the back
of the truck as well (step 508). The position and speed estimate is
then propagated in time to the position expected for the next radar
sample in step 510. That is, the logic estimates the expected
position of the back of the truck at the time the next radar sample
is expected. This is a relatively simple matter since the radar
samples are provided at regular intervals so the timing of the next
expected sample is easy to determine. For example, if the radar
sample rate is 20 Hz, the next sample can be expected to occur 0.05
seconds after the last sample. If the front and rear vehicles are
traveling at exactly the same velocity and both vehicles are
traveling in the same direction, than the "expected" position of
the back of the front vehicle would be exactly the same as the last
detected position of the back of the front vehicle. However, often
vehicles will be traveling at slightly different speeds and
possibly in slightly different directions if one of the vehicles is
turned or turning slightly relative to the other. For example,
using a simple example, if the trailing vehicle is moving in
exactly the same direction as the lead vehicle at a constant
velocity of 1.00 meters per second faster than the lead vehicle,
then the back of the lead vehicle would be expected to be 5 cm
closer to the lead vehicle at the time the next radar sample is
taken (0.05 seconds after the last sample was taken). Simple
trigonometry may be used to determine the expected position if the
vehicles are turned or turning slightly with respect to one
another. Of course, any number of other relevant variables that are
known to or obtainable by the radar system controller can be
considered in the calculation of the expected position and speed to
further improve the estimates. These might include the respective
accelerations (measured or estimated) of the vehicles, the
respective directions of travel and/or rates of turn of the two
vehicles, etc. Factors that may influence the velocity,
acceleration or rate of turn of the vehicles such as the respective
vehicles torque requests, the current grade, the vehicle weights,
etc. may also be used to further refine the estimate.
[0108] In addition to propagating the position estimate in time,
the uncertainty estimate is updated as represented by block 512 as
described in more detail below.
[0109] After the position estimate has been propagated in time and
the uncertainty estimate has been updated, the process repeats for
the next sample as represented in the flow chart of FIG. 7 by
returning to step 502 where the next radar scene sample is
received. The propagation of the estimated position in time is
particularly useful in step 504 which utilizes the then current
estimate of the position of the back of the lead vehicle to
determine whether a match occurs. The current estimate of the
position of the lead vehicle can be expected to (indeed likely
will) change over time. For each radar sample, the then current
best estimate of the position of the back of front vehicle may be
used which helps ensure that the partner vehicle is accurately
tracked.
[0110] As suggested above, the platoon system preferably utilizes
multiple independent or partially-independent mechanisms for
tracking the position and speed, of the respective vehicles. For
example, as discussed above, the platoon controller may have access
to GPS position data which provides an independent mechanism for
determining the relative positions of the platooning vehicles. The
platoon controller may also have access to wheel speed data which
provides an alternative mechanism for determining the respective
speeds, and thus the relative speed of the platoon partners. Such
data for the host vehicle is available from the host vehicle
sensors. Data for the partner vehicles is available over the
communications link (e.g. the DSRC link, a cellular link or any
other available communication method).
[0111] Each time that a new GPS position estimates are received (as
represented by box 520 in FIG. 7), the radar tracking position and
speed estimate is updated using the current GPS position estimate
(step 523), and that updated position and speed estimate is
propagated in time to the expected receipt of the next radar sample
as represented by step 510. In parallel, each time that new wheel
speed estimates are received (as represented by box 530 in FIG. 7),
the radar tracking position and speed estimate is updated using the
current wheel speed estimates (step 533), and that updated position
and speed estimate is propagated in time to the expected receipt of
the next radar sample as represented by step 510. Similarly, each
time new inertial measurements such as yaw rates, vehicle
orientation (heading), vehicle pitch and/or vehicle roll are
received (as represented by box 540), the radar tracking position
and speed estimate s updated using the current inertial
measurements (step 542).
[0112] The GPS position, wheel speed and inertial measurements are
preferably updated on a relatively rapid basis--which is often
(although not necessarily) more frequent than the radar samples. By
way of example, GPS update frequencies in the range of 25 to 500
Hz, as for example 50 Hz have been found to work well for open road
platoon control applications. Similar wheel speed and inertial
measurement update frequencies have also been found to work
well--although there is no need to update the GPS positions, wheel
speed and/or inertial measurements at the same sample rate as each
other, or at the same sample rate as the radar unit.
[0113] In the embodiment shown, the updates from the radar unit,
the GPS sensors, the wheel speed sensor and inertial measurements
are handled asynchronously as they are received. Although not
required, this is useful to help ensure that the latest sensor
inputs are utilized in estimating the expected relative positions
and speeds of the platooning vehicles at the time the next radar
unit scene sample is received. This is contrasted with a system in
which the wheel speed sensor and GPS sensor information is updated
once each sample of the radar unit. Although synchronous updates
can also work well, the use of asynchronous updates tends to
improve the accuracy of the estimates because various sensor inputs
can be updated more frequently than the radar unit sampling
rate.
[0114] Although the different types of measurements do not need to
be synchronized with one another, the same types of measurements on
the different trucks are preferably synchronized in time. That is,
GPS position measurements on the front truck are preferably
synchronized in time with GPS position measurements on the back
truck so that the relative positions of the trucks can be
determined at a particular instant in time. Similarly, the wheel
speed measurements on the front truck are preferably synchronized
in time with wheel speed measurements on the back truck so that the
relative speeds of the trucks can be determined at a particular
instant in time. The various inertial measurements are also
preferably synchronized with each other as well.
[0115] It should be appreciated that it is relatively simple to
coordinate the timing of the various measurements between vehicles
because GPS is used and the vehicles communicate with one another
over the communications link. As is well known, the GPS system
provides very accurate global timing signals. Thus, the clocks used
for the platoon partners can be synchronized with the GPS signals
and the various measurements (e.g. GPS position measurements, wheel
speed measurements, inertial measurements, etc.) can therefore be
instructed to occur at specific synchronized times on the
respective trucks. Each measurement may also be accompanied by a
timestamp that indicates when the measurement was taken so that the
synchronization of the measurements can be verified (or accounted
for if similar sensor measurements are not synchronized between
vehicles).
[0116] The propagation of the estimated position in time is
particularly useful in step 504 which utilizes the then current
estimate of the position of the back of the lead vehicle to
determine whether any of the received radar sample object points
(targets) match the expected position of the back of the partner
vehicle. It should be appreciated that there may be times when no
radar sample targets match the expected position of the back of the
partner vehicle as represented by the "no" branch from decision
504. In such cases the radar system controller still propagates the
position estimate in time (step 510) so that the position estimate
is updated for the next radar sample based on the other information
the controller has. Such other information includes the then
current estimates and may be further updated based on inputs from
other systems (e.g., the GPS or wheel speed sensor) as previously
discussed.
[0117] There are some operational circumstances where one or more
measurements might be expected to be suspect. For example, when a
host vehicle is shaken unusually hard--as may occur when a wheel
runs over a pothole or encounters other unusual roughness in the
road--the radar unit will be shaken accordingly and any radar
measurement samples taken at that instant are less likely to be
accurate and/or useful to the model. Other sensors such as the
wheel speed and inertial measurement sensor are less likely to be
accurate at such times as well. In another example, when the lead
truck is aggressively braking it is more likely that its trailer
will move back and forth more than usual which again suggests that
any radar samples taken during such braking are less likely to be
useful for predicting the future position of the back of the
trailer. When the controller detects, or is informed, that an event
is occurring that makes the measurements of any particular sensor
suspect, the measurements from such sensor(s) can safely be ignored
in the context of the position estimate. In such circumstances
inputs from other sensors deemed more reliable (if any) may
continue to be used to update the position model and the position
estimate may continue to be propagated in time for each subsequent
sample. The uncertainty associated with position estimate can be
expected to increase slightly with each ignored sample, which has
the effect of increasing the variation from the estimated position
of the back of the partner vehicle that would be tolerated when
determining whether there is a target that matches the expected
position of the back of the partner vehicle.
[0118] The position model described above is relatively simple in
that it utilizes a relatively small set of measured inputs
including (1) the received radar scenes (which show the relative
position and relative velocity of detected objects); (2) measured
GPS positions of the platoon partners (which can be used to
determine their relative positions); (3) measured wheel speeds of
the platoon partners (which can be used to determine their relative
speeds); and (4) measured yaw rate and orientation. In other
embodiments, when different or additional types of sensor
information is available to the radar controller, the position
model can be adapted to utilize whatever relevant information is
available to it in the position estimates. For example, if the
pitch or roll of the vehicles are available, the position model can
incorporate such measurements into the position estimates. The roll
can be useful because on trucks the GPS antennas tend to be located
on top of the cabs at locations over 4 meters above the ground
(e.g. 14-15 feet). At such heights, even relatively small tilting
in the roll direction can cause the reported position of the
respective vehicles to vary significantly. The pitch can be useful
for similar reasons. For example, with a platooning gap of 15
meters, a difference in pitch of just .+-.2 degrees can result in a
difference of a meter in the apparent or detected height of an
object. At further distances and/or larger pitch variations, those
differences are amplified. Since many radar units used in
platooning systems have relatively narrow views this can lead to
expected objects not being detected, or detected objects being
discarded, because they are further from the estimated position
than expected when pitch is not considered. Similarly, if
longitudinal and/or angular accelerations are available, the
position model can incorporate the acceleration measurements into
the position estimates.
[0119] In embodiments in which the relative positioning and/or
speed and/or orientation of the vehicles can relatively accurately
be measured using other systems such as LIDAR, sonar, other time of
flight distance sensors, sensors configured to receive a signal
transmitted from another vehicle, cameras, stereo cameras or other
appropriate technologies, those measurements can be incorporated
into the position model in addition to, or in place of, the GPS,
wheel speed and inertial measurements.
[0120] In some embodiments, the position model can be considerably
more sophisticated using inputs such a torque requests, braking
signals and/or other operational information about the respective
platoon partners to further refine the predicted position at the
time of the next radar sample.
[0121] In the primary described embodiment the radar sample object
points are compared to the estimated (expected) position and
relative speed of the back of the partner vehicle. In other
embodiments, more or fewer parameters can be compared to identify a
match. For example, in some embodiments matches (or lack thereof)
may be based on matching the expected position of the partner
vehicle rather than position and speed/velocity. If the radar unit
is capable of reliably reporting other information such as
acceleration, rates of lateral movement, etc., then such
information can also be compared to corresponding estimates as part
of the match identification 504.
[0122] A significant advantage of the described approach is that
the relative position and velocity estimates can reliably continue
even when the back of the platoon partner is outside the view of
the radar unit--as may sometimes be the case when the lead vehicle
changes to a different lane, an interloper cuts in between the
platooning vehicles, or a transitory fault occurs with the radar
unit. With such tracking, radar identification of the platoon
partner can more easily be reestablished when the back of the
platoon partner comes back into the radar unit's view. As will be
appreciated by those familiar with the art, this is very different
than adaptive cruise control systems that utilize radar only to
track the distance to the vehicle directly in front of the host
vehicle--regardless of who that leading vehicle may be.
[0123] It is noted that the histogram and/or mean shift clusters
described above with respect to FIG. 5 can be used as another check
to verify that the correct vehicle is being tracked by the radar
unit or to provide a reference point when some, but not all of the
truck is within the radar unit's field of view.
[0124] A noteworthy feature of the method described with respect to
FIG. 7 is that the same algorithm(s) can be used to estimate the
relative position/velocity of the partner vehicle during the
initial radar identification of the partner vehicle as described
above with respect to FIG. 2. In that situation, the radar tracker
116/600 would not have a good estimate of the position of the back
of the partner vehicle. As such, no target would match the expected
position of the back of the partner vehicle at decision point 504
so no measured position would be reported to the gap controller and
the radar unit's measurements would not be used to update the
position and speed estimates--thereby following the "no" branch
from decision point 504 which causes steps 506 and 508 to be
skipped. However, the other available sensors, including the GPS
sensors 131, the wheel speed sensors 132 and inertial measurement
sensors 134 all provide their respective measurements, which
provides a reasonable estimate of the position of the vehicle
suitable for use in the initial identification of the partner
vehicle.
Kalman Filtering
[0125] The method described with respect to FIG. 7 can be
implemented using a variety of techniques. One presently preferred
embodiment that works particularly well utilizes Kalman Filtering.
As used herein, the phrase Kalman filtering is intended to
encompass linear quadratic estimation (LQE) as well as extensions
and generalizations of LQE such as the extended Kalman filter and
the unscented Kalman filter which are designed to work with
nonlinear systems. As will be understood by those familiar with
Kalman filtering in general, Kalman filtering uses a series of
measurements observed over time containing noise and other
inaccuracies and produces estimates of unknown variables that tend
to be more precise than those based on a single measurement alone.
The Kalman filter keeps track of the estimated state of the system
and the variance or uncertainty of the estimate. This is
particularly well suited for estimating the position, speed and
other state information related to gap control because of the
errors inherent is some of the measurements and the potential
unavailability at times of some of the desired measurement
samples.
[0126] The state variables used in the Kalman filter may vary
widely with the nature of the model used. One particular state
array (X) suitable for use in some of the described embodiments
that involve a pair of platooning tractor-trailer trucks
includes:
[0127] (1) the longitudinal position of the center of the rear
axles of the front truck relative to the center of the rear axles
of the back truck (x);
[0128] (2) the lateral position of the center of the rear axle of
the front truck relative to the center of the rear axles of the
back truck (y);
[0129] (3) the heading of the front truck relative to the heading
of the trailing truck (.chi.);
[0130] (4) the speed of the lead vehicle (v.sub.1); and
[0131] (5) the speed of the trailing vehicle (v.sub.2).
[0132] This can be represented mathematically as follows:
X = [ x y .chi. v 1 v 2 ] ##EQU00001##
[0133] The estimated state at the time of the next radar sample
(X.sub.k+1) is a function of the previous state (X.sub.k) and a
covariance matrix (P.sub.k) indicative of the level of uncertainty
in the measurements. A covariance matrix corresponding to the state
array (X) represented above is illustrated in FIG. 8. As will be
understood by those familiar with Kalman filtering in general, the
estimated state at the time of the next radar sample (X.sub.k+1) is
equal to the product of a state transition model (A) and the
previous state (X.sub.k) plus the product of a control input model
(B) and any modeled inputs (u.sub.k-1). This can be represented
mathematically as follows.
X.sub.k+1=AX.sub.kBu.sub.k
[0134] One particular control input array (U) includes:
[0135] (1) the yaw rate of the front vehicle (.psi..sub.1); and
[0136] (2) the yaw rate of the rear vehicle (.psi..sub.2)
[0137] This can be represented mathematically as follows:
U = [ .psi.1 .psi.2 ] ##EQU00002##
[0138] Although specific state and modeled input arrays are
illustrated, it should be appreciated that the specific state and
control input variables used in any particular implementation may
vary widely based on the nature of the estimation model used.
[0139] Kalman filtering is particularly well adapted to making the
types of position and velocity estimations useful in the techniques
described herein. Although Kalman filtering works particularly
well, it should be appreciated that other state/space estimation
algorithms, such as Particle Filtering, etc. can be used in
alternative embodiments.
[0140] One of the reasons that Kalman filtering works well is that
most of the measurements, including the GPS measurements, the radar
measurements, the wheel speed measurements and the inertial
measurements tend to be subject to varying measurement errors. For
example, it is not uncommon for any particular GPS measurement to
be off by more than a meter. The covariance matrix (P.sub.k)
quantifies the statistical variation (error) observed in the
measurements and utilizes that knowledge to improve the quality of
the position and speed estimates.
Integrating Other Information into Sensor Data Verification
[0141] In the embodiments described above, information about the
state of the partner vehicle that is received from the partner
vehicle is used by the host to help verify or confirm that data
from a sensor on the host vehicle that is believed to measure a
characteristic of the partner vehicle is actually representative of
the partner vehicle. For example, in some of the described
embodiments, information from a lead vehicle about its position,
speed, orientation etc. is used by a radar scene processor on the
trailing vehicle to predict an expected position and speed of the
lead vehicle. Those predictions are then used to help determine
which (if any) of the detected radar objects correspond to the lead
vehicle. The state information received from the lead vehicle may
be a measured value (such as a measure wheel speed) or a predicted
value (such as a predicted speed) which may be even more reliable
in circumstances in which the parameter (e.g., speed) is
changing.
[0142] It should be appreciated that a wide variety of other
information/data received from the partner vehicle can additionally
or alternatively be used to further help with such verification.
This can include other partner vehicle state information such as
the partner vehicle's: current torque request; braking status
(including the status of the foundation brakes, a retarder, engine
braking and/or any other braking device in the context of larger
trucks); or steering angle. The information can also include a
status indicator such as an indication that a blinker, the hazard
lights, the taillights or other lights are on. It can also include
qualitative information about the partner vehicle such as its radar
signature, or its visual appearance (e.g. its color, a identifying
marker, or some other feature or characteristic that can be readily
identified by one of the controllers on the host vehicle). It can
also include information about an intended or expected action--such
as notification that the lead vehicle is about to change lanes,
will take the next exit or turn at the next intersection.
[0143] In some circumstances, the host vehicle may request that the
partner vehicle take specific actions to help with such
identification. The nature of such a request may vary widely--for
example, the rear truck may request that the lead truck turn on
specific lights, switch lanes, accelerate or decelerate to a
specific speed, honk its horn, etc.
[0144] Additionally, it should be appreciated that additional
information about the partner vehicle can also be obtained from a
third vehicle, a larger mesh of vehicles or from another external
source. For example a third vehicle travelling in parallel with the
platoon partners may have measured the position, velocity and/or
other characteristics of the partner vehicle and that information
can be used as another independent check. In another example, a
network operations center (NOC) in communication with both platoon
partners may know the intended route and communicate that route, or
more short term directions to the host vehicle as appropriate. In
other circumstances information from the partner vehicle may be
transmitted via an intermediary such as a third vehicle, a NOC,
etc. Any of this type of data can be useful--and some of the
information may be particularly helpful in circumstance in which
communications between the vehicles is temporarily lost.
[0145] Although only a few embodiments of the inventions have been
described in detail, it should be appreciated that the inventions
may be implemented in many other forms without departing from the
spirit or scope of the invention. The inventions have been
described primarily in the context of a pair of trucks platooning
with a forward facing radar unit being located at the front of the
trailing truck. However, it should be appreciated that the same
concepts can be applied to any types of vehicles operating in any
type of connected vehicle applications, regardless of where the
radar unit is located on the vehicle and/or the direction (or
directions) that the radar unit(s) interrogates. Thus, for example,
a backward facing radar unit on a lead vehicle can be used to
identify and/or track following vehicles using radar in
substantially the same manner as described. Similarly if
omni-directional radar is used, similar approaches can be used to
identify and/or track other vehicles using radar regardless of
their position relative to the host vehicle.
[0146] As suggested above, the described radar based vehicle
identification and tracking can be used in any type of connected
vehicle application in which independent information about the
position and/or velocity of one or more other vehicles is known or
available to the unit interpreting the radar data. Thus, for
example, the described techniques are particularly well suited for
use in convoying systems involving more than two vehicles. Also,
the described techniques are very well adapted for use in
autonomous vehicle traffic flow applications where knowledge about
the intentions of other specific vehicles is deemed important.
Indeed, this is expected to be an important application of the
inventions with the growth of the autonomous and connected vehicle
markets.
[0147] The inventions have been described primarily in the context
of identifying and tracking other vehicles using commercially
available radar units designed for use in driving automation
systems. Such units are typically designed to analyze the received
radar energy and identify objects that are believed to the radar
manufacturer to be relevant. Although the described inventions work
well with such units, they are not so constrained. Rather, both the
vehicle identification and vehicle tracking processes are well
suited for use with radar units that don't filter the response as
much and report the reflected radar signal intensities in a more
general way rather than attempting to identify particular objects.
In particular, the statistical nature of the radar return binning
and the back of vehicle detection are quite well suited for using
radar data provided in other forms such as intensity/location.
Furthermore, the invention is not limited to distance measurement
systems using electromagnetic energy in the frequency range of
radar. Rather, it should be appreciated that the same target
vehicle identification and/or tracking techniques may readily be
used in conjunction with other electromagnetic energy based
distance measuring technologies such as LIDAR which utilize
electromagnetic energy in different frequency ranges, sound based
distance measurement (e.g., sonar, ultrasound, etc.) or various
time of flight based distance measuring systems. The described
techniques can also be used in conjunction with distance measuring
techniques using cameras or stereo cameras, beacon based
technologies in which the sensor measures a beacon signal
transmitted from the partner vehicle and/or other technologies.
[0148] In some implementations, the platooning vehicles may have
mechanisms such as transponders suitable for identifying themselves
to the radar unit. When available, information from such devices
can be used to further assist with the identification and tracking
of the platoon partner.
[0149] Therefore, the present embodiments should be considered
illustrative and not restrictive and the invention is not to be
limited to the details given herein, but may be modified within the
scope and equivalents of the appended claims.
* * * * *