U.S. patent application number 12/947620 was filed with the patent office on 2011-05-19 for vehicle control system.
Invention is credited to Andrew John MacDonald, Campbell Robert Morrison, David Robert Reeve.
Application Number | 20110118938 12/947620 |
Document ID | / |
Family ID | 39588075 |
Filed Date | 2011-05-19 |
United States Patent
Application |
20110118938 |
Kind Code |
A1 |
MacDonald; Andrew John ; et
al. |
May 19, 2011 |
VEHICLE CONTROL SYSTEM
Abstract
A vehicle control system having a controller and a spatial
database adapted to provide spatial data to the controller at
control speed. The spatial data provided from the spatial database
to the controller can be any kind of data or information that has
some relationship or association with "real world" geographical
location, or if it is stored somehow with reference to geographical
location. The spatial data received by the controller from the
database forms at least part of the control inputs that the
controller operates on to control the vehicle. The fact that the
controller operates directly on information that is inherently
associated with "real world" geographic location represents a
change in thinking compared with existing vehicle control systems.
In particular, it means that the control system of the present
invention "thinks" directly in terms of spatial location. A vehicle
control system in accordance with one particular embodiment of the
invention comprises a task path generator, a spatial database, at
least one external spatial data receiver, a vehicle attitude
compensation module, a position error generator, a controller, and
actuators to control the vehicle.
Inventors: |
MacDonald; Andrew John;
(Graceville, AU) ; Reeve; David Robert; (Chapel
Hill, AU) ; Morrison; Campbell Robert; (Corinda,
AU) |
Family ID: |
39588075 |
Appl. No.: |
12/947620 |
Filed: |
November 16, 2010 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
11620388 |
Jan 5, 2007 |
7835832 |
|
|
12947620 |
|
|
|
|
Current U.S.
Class: |
701/41 ; 701/1;
701/93 |
Current CPC
Class: |
G01S 5/0252 20130101;
A01B 69/008 20130101; G01C 21/16 20130101; G05D 1/0223 20130101;
G01S 19/03 20130101; G05D 1/027 20130101; G01S 17/86 20200101; G05D
1/0255 20130101; G05D 2201/0201 20130101; G05D 1/0278 20130101;
A01B 79/005 20130101; G05D 1/024 20130101; G05D 1/0259
20130101 |
Class at
Publication: |
701/41 ; 701/1;
701/93 |
International
Class: |
G05D 1/10 20060101
G05D001/10; G05D 1/00 20060101 G05D001/00 |
Claims
1. A control system as claimed in claim 1, wherein data is arranged
within the database in accordance with a hash table which relates
the memory allocations for the different items of spatial data
within the database to corresponding indices.
2. A control system as claimed in claim 30, wherein the indices for
the different items of data are determined according to the spatial
location to which the respective items of data pertain.
3. A control system as claimed in claim 31, wherein a spatial hash
key algorithm is used to generate the indices.
4. A control system as claimed in claim 32, wherein the spatial
hash key algorithm operates so that data pertaining to spatial
locations which are close to each other receive closely related
indices.
5. A control system as claimed in claim 33, wherein the spatial
hash key algorithm uses bitwise interleaving.
6. A control system as claimed in claim 34, wherein the spatial
hash key algorithm uses double-precision floating-point
numbers.
7. A control system as claimed in claim 1, wherein the control
system is adapted to receive data from at least one external
source.
8. A control system as claimed in claim 36, wherein the control
system is adapted to receive data from at least one external source
at control speed.
9. A control system as claimed in claim 36, wherein the data
received from the at least one external source is used to generate
estimates of the vehicle's pose.
10. A control system as claimed in claim 36, wherein the at least
one external data source includes GPS.
11. A control system as claimed in claim 39 wherein the GPS is
supplemented by a SBAS.
12. A control system as claimed in claim 36, wherein the at least
one external data source includes an INS.
13. A control system as claimed in claim 41 wherein the INS
includes one or more rate gyros, accelerometers or a combination of
both.
14. A control system as claimed in claim 39, wherein the GPS is
supplemented by a GBAS.
15. A control system as claimed in claim 36, wherein the at least
one external data source includes machine vision and/or image
analysis.
16. A control system as claimed in claim 36, wherein the at least
one external data source includes LIDAR.
17. A control system as claimed in claim 36, wherein the at least
one external data source includes a magnetometer.
18. A control system as claimed in claim 36, wherein the at least
one external data source includes a tilt sensor.
19. A control system as claimed in claim 36, wherein the at least
one external data source includes ultrasonic range and direction
finding.
20. A control system as claimed in claim 36, wherein a filter is
used to obtain a statistically optimised estimate of the state of
the vehicle.
21. A control system as claimed in claim 48, wherein the filter
utilises the data from the one or more external data sources to
obtain the optimised estimate.
22. A control system as claimed in claim 49, wherein the filter is
a Kalman filter.
23. A control system as claimed in claim 1, including means for
interpreting a user-defined path and converting the said
user-defined path into a plurality of points representing the
user-defined path.
24. A control system as claimed in claim 51, including means for
generating a desired position, heading and instantaneous radius of
curvature for the vehicle at each point on the user-defined
path.
25. A control system as claimed in claim 52, wherein the points
representing the user-defined path and the desired position,
heading and instantaneous radius of curvature for the vehicle are
entered into the spatial database.
26. A control system as claimed in claim 53, including means for
calculating an error term relating to the difference between the
vehicle's actual position, heading and instantaneous radius of
curvature and the vehicle's desired position, heading and
instantaneous radius of curvature.
27. A control system as claimed in claim 54, wherein the controller
uses the error term to generate a control signal for controlling
the vehicle.
28. A method for controlling a vehicle comprising entering spatial
data relating to a region to be traversed by the vehicle into a
spatial database, providing spatial data from the spatial database
to a controller at control speed to control the vehicle as the
vehicle traverses the region, and entering updated spatial data
into the spatial database as the vehicle traverses the region.
29. A method for controlling spatial data from the spatial database
to a controller at control speed to control the vehicle as the
vehicle traverses the region, and updated spatial data being
received into the spatial database as the vehicle traverses the
region.
Description
CROSS-REFERENCE TO RELATED APPLICATION
[0001] This application is a continuation of and claims the benefit
of U.S. patent application Ser. No. 11/620,388, filed Jan. 5, 2007,
entitled "Vehicle Control System," now U.S. Pat. No. 7,835,832,
issued Nov. 16, 2010, which is incorporated herein by
reference.
FIELD OF THE INVENTION
[0002] The present invention relates to a control system for
controlling the direction of travel of a vehicle, and in particular
to a control system having an embedded spatial database. The
control system of the present invention may also be used to control
other aspects of a vehicle's motion, such as speed or acceleration.
Furthermore, in the case of agricultural vehicles and the like, the
present control system may be used to control yet other aspects of
the vehicle's operation, such as the application of agricultural
chemicals at desired locations (including at desired application
rates), or the engagement and/or mode of operation of agricultural
implements (e.g. ploughs, harvesters etc) at desired locations,
etc.
[0003] For convenience, the invention will be described mainly with
reference to agricultural vehicles and like moving agricultural
machinery. However, it will be clearly understood that the
invention is not limited to agricultural applications and it may
equally be applied to vehicles and other moving machinery in other
areas.
BACKGROUND
[0004] A number of control systems have previously been devised for
controlling the steering of agricultural vehicles. These systems
are generally used on vehicles such as tractors (including tractors
with towed tools or other implements), harvesters, headers and the
like which operate in large fields. These vehicles generally move
along predetermined trajectories ("paths") throughout the field. In
general, a wayline is entered into the control system and
subsequent paths are calculated based on the wayline. If the
vehicle deviates from the path as it moves, the controller causes
the vehicle to steer back towards and onto the path as described
below.
[0005] As the vehicle moves along the predetermined path
trajectory, it uses various means such as signals produced by GPS
(global positioning system) or INS (inertial navigation system) to
identify if the vehicle deviates from the desired path trajectory.
If the vehicle deviates, the extent of the deviation (i.e. the
difference between the actual curvature of the vehicle's trajectory
and the desired curvature, its actual compass heading compared with
the desired compass heading, and the distance the vehicle is
displaced laterally from the desired path) is expressed in the form
of an error, and this error is fed back into the control system and
used to steer the vehicle back onto the desired path.
[0006] A problem with previous vehicle control systems is that they
are inherently "one-dimensional" or "linear" in nature. This means
that, at a fundamental level, the controller operates by "knowing"
the path that the vehicle is required to traverse, and "knowing"
where the vehicle is located on that path (i.e. how far along the
path the vehicle has moved) at a given time. However, the
controller does not "know" where the vehicle is actually located in
space. This is despite the fact that the controller may often
progressively receive information containing the vehicle's spatial
location, for example from the GPS/INS signals. In current
controllers, the GPS/INS signals are used primarily to determine
when the vehicle deviates from the path (i.e. to calculate the
error) rather than for the primary purpose of determining the
vehicle's actual position in space. Hence, at a fundamental level,
the controller only "knows" the geometry of the path and how far
the vehicle has moved along the path.
[0007] Therefore, with current controllers, if it is desired to
know the actual spatial position of the vehicle, this must be
calculated from the known geometry of the path and the known
distance the vehicle has moved along that path. This calculation
can be computationally expensive and difficult to implement in
practice, particularly for curved, piecewise, broken or other
complex path trajectories.
[0008] By way of example, it will be appreciated that one form of
common path trajectory that agricultural vehicles are often
required to traverse in fields is made up of a number of (usually
parallel) path segments or "swaths" (these are sometimes also
referred to as "rows"). Thus, the vehicle typically moves along one
swath, harvesting or ploughing as it goes, and it then turns around
and moves back along an adjacent parallel swath, harvesting or
ploughing in the opposite direction. The adjacent swath will
generally be spaced from the first swath sufficiently closely that
no part of the field or crop is missed between the swaths, but also
sufficiently apart so that there is not an unnecessary overlap
region (i.e. a region between the swaths that gets ploughed or
harvested on both passes). In general, the distance between the
mid-lines of each respective swath is determined with reference to
the width of the vehicle (i.e. the width of the plough, harvester
or possibly the tool being towed by the vehicle).
[0009] In cases where paths comprising a series of parallel swaths
are used, the first swath will often be used as a reference swath
or "wayline". In general, the geometry of the wayline in space will
be entered into the control system along with the vehicle or
implement width, and this is used to calculate the required spacing
(and hence trajectory) for each of the adjacent parallel swaths.
However, with most existing control systems, the controller is only
able to control the steering of the vehicle as it proceeds along
each of the swaths. It is much harder to control the steering of
the vehicle as it turns around between one swath and the next.
Therefore, whilst the spatial geometry of the respective swaths may
have been calculated, from the control system's point of view at
any given time it only "knows" that it is on the nth swath
(numbered from the wayline) and that it has been moving along that
swath for a known amount of time with known speed (i.e. it knows
that the vehicle is a certain distance along the nth swath).
However, at a fundamental level, the control system does not
inherently know where the vehicle is consequently located in space
or the spatial relationship between each swath. A graphical
representation of the difference between the vehicle's actual
spatial location and what the control system "sees" is given in
FIG. 1.
[0010] The "one-dimensional" or "linear" nature of existing control
systems also causes other difficulties. One example is in relation
to obstacle avoidance. In most agricultural applications, the
positions of obstacles (e.g. fences, trees, immovable rocks, creeks
etc) are known according to their "real-world" spatial location.
The spatial location may be known according to global latitude and
longitude coordinates (e.g. as provided by GPS), or alternatively
the location may be known relative to a fixed point of known
location (this is generally a point in or near the field used to
define the origin of a coordinate system for the field). However,
as current control systems only recognise where the vehicle is
located along the path, not where the vehicle is actually located
in space, the control system itself is therefore unable to
recognise whether the location of the obstacle coincides with the
trajectory of the path, and hence whether there may be a
collision.
[0011] Consequently, with current control systems, it may be
necessary for a number of separate modules to be provided, in
addition to the primary control module, if automatic obstacle
avoidance (i.e. obstacle avoidance without the need for
intervention by the driver of the vehicle) is to be achieved. In
these cases, one of the modules would be a collision detection
module for calculating the geometry and trajectory of a section of
the path a short distance ahead of the vehicle in terms of "real
world" spatial coordinates and for determining whether any of the
points along that section of path will coincide with the location
of an obstacle. If the collision detection module identifies that
the section of path is likely to pass through an obstacle (meaning
that there would be a collision if the vehicle continued along that
path), then a further module may be required to determine an
alternative trajectory for (at least) the section of the path
proximate the obstacle. Yet a further module may then be required
to determine how best to steer the vehicle from the alternative
trajectory back onto the original path after the vehicle has moved
past the obstacle. This multi-modular control system structure is
complicated and can lead to computational inefficiencies because
the different modules may each perform many of the same geometric
calculations for their own respective purposes, separately from one
another, leading to "doubling up" and unnecessary computation.
Also, with this modular control system structure, control of the
vehicle generally passes from one module to another as described
above, but determining when one module should take over from
another creates significant difficulties in terms of both system
implementation and maintenance
[0012] Another problem associated with the "one-dimensional" nature
of existing control systems is their inherent inflexibility and
unadaptability. For example, in practice, if the vehicle deviates
from the desired path for some reason, it may be preferable for
subsequent paths (swaths) to also include a similarly shaped
deviation so that the paths remain substantially parallel along
their length (or tangentially parallel and consistently spaced in
the case of curved sections of path). If the vehicle is, for
example, a harvester or a plough, then keeping the paths parallel
in this way may help to prevent portions of the field from being
missed, or from being harvested/ploughed multiple times (by passing
over the same portion of field on multiple passes). Even with the
modular control system structures described above, it is often
difficult to determine the geometry of the deviated path portion in
terms of "real world" coordinates, and even if this can be done, it
is also difficult to adjust subsequent path geometries to
correspond to the deviation from the predetermined path trajectory
that was originally entered.
[0013] As a further example of the inherent inflexibility and
unadaptability of current "one-dimensional" control systems, it is
illustrative to consider the situation where an obstacle is located
near the end of one swath such that it would be quicker and more
efficient to simply move on to an adjacent swath located nearby
rather than wasting time trying to go around the obstacle to finish
the first swath before moving on to the adjacent swath. Current
"one-dimensional" control systems are not able to recognise that it
would be more efficient to move on. This is because the control
system only knows where the vehicle is along its current path (e.g.
close to the end of the swath), and if a modular control systems is
used, that module may also recognise that it is approaching the
obstacle. The control system does not know where the vehicle is
actually located in space, and therefore it cannot recognise that
the beginning of the next swath is actually located nearby--it
simply does not know where the next swath is (or indeed where the
current swath is in space). Therefore, current control systems
cannot easily recognise when it would be better to change paths (at
least without intervention from the vehicle's driver), as this
example illustrates. Nor is the current "one-dimensional" structure
inherently adapted to enable the control systems to automatically
(i.e. autonomously without assistance from the driver) determine
and guide the vehicle along an efficient trajectory between
swaths.
[0014] It will be clearly appreciated that any reference herein to
background material or a prior publication is not to be understood
as an admission that any background material, prior publication or
combination thereof forms part of the common general knowledge in
the field, or is otherwise admissible prior art, whether in
Australia or any other country.
DESCRIPTION OF THE INVENTION
[0015] It is an objective of the present invention to provide a
vehicle control system having an embedded spatial database that may
at least partially ameliorate one or more of the above-mentioned
difficulties, or which may provide a useful or commercial
alternative to existing control systems.
[0016] Accordingly, in a first broad form the present invention
resides in a vehicle control system having a controller and a
spatial database adapted to provide spatial data to the controller
at control speed.
[0017] In another broad form, the invention resides in a control
system for controlling a vehicle within a region to be traversed,
the control system comprising
[0018] a spatial database containing spatial data,
[0019] a controller adapted to receive spatial data from the
spatial database at control speed,
[0020] the control system being adapted to receive spatial data
from the controller and/or an external source,
[0021] the controller using the spatial data for controlling the
vehicle.
[0022] In a further broad form, a control system is provided for
steering a vehicle within a region to be traversed, the control
system comprising
[0023] a spatial database containing spatial data,
[0024] a controller adapted to receive spatial data from the
spatial database at control speed,
[0025] the controller being adapted to control the steering of the
vehicle,
[0026] the spatial database being adapted to receive updated
spatial data from the controller and/or an external source,
[0027] the updated spatial data relating to the vehicle and/or an
implement associated with and proximate the vehicle and/or at least
a portion of the region proximate the vehicle.
[0028] In agricultural applications, the region to be traversed by
the vehicle will generally be the field that is to be ploughed,
harvested, etc, and the invention will be described generally with
reference to agricultural vehicles operating in fields. However, no
limitation is meant in this regard, and the region to be traversed
by the vehicle may take a range of other forms in different
applications. For example, in automotive applications the region to
be traversed by the vehicle might comprise roadways located in a
particular geographical area. Alternatively, in mining applications
the region could comprise the vehicle navigable regions of the
mine. In underground mining, this could include the various levels
of the mine located vertically above and below one another at
different relative levels (depths). Furthermore, the control system
of the present invention could be applied to vehicles that operate
on airport tarmacs, in which case the region to be traversed by the
vehicle might be the tarmac, or a portion thereof. From these
examples, the person skilled in the art will appreciate the breadth
of other applications that are possible.
[0029] The control system of the present invention includes a
spatial database that contains spatial data. The spatial database
may also be adapted to receive spatial data including updated
spatial data, and to provide spatial data to other components of
the control system. In general, data may be characterised as
"spatial" if it has some relationship or association with "real
world" geographical location, or if it is stored somehow with
reference to geographical location. Some illustrative examples of
the kinds of spatial data that may be stored within the database
include (but are not limited to) coordinate points describing the
location of an object (e.g. a rock or tree) in terms of the
object's "real world" geographical location in a field, the
coordinate points for a geographical location itself, information
regarding a "state" of the vehicle (e.g. its speed, "pose"
(position and orientation) or even fuel level) at a particular
geographical location, a time when the vehicle was at a particular
geographical location, or a command to the vehicle to change its
trajectory or mode of equipment (e.g. plough) operation if or when
it reaches a certain geographical location. These examples
illustrate that any data or information that has an association
with geographical location, or which is stored with reference to
geographical location, can constitute "spatial data". For the
remainder of this specification, the terms "spatial data" and
"spatial information" will be used interchangeably. References
simply to "data" or "information" will generally also carry a
similar meaning, and references simply to the "database" will be to
the spatial database, unless the context requires otherwise.
Typically, the spatial database is an electronic database stored in
a memory device, such as, for example, a RAM, as discussed in more
detail below.
[0030] Spatial data may be stored within the database according to
any convenient coordinate system, including (but not limited to)
cartesian (or projected) coordinates, polar coordinates,
cylindrical coordinates, spherical coordinates,
latitude/longitude/altitude etc. The coordinate system may also be
"global" in the sense of the location references provided by GPS,
or "local" coordinates such as those defined with respect to a
local origin and reference orientation. The coordinates may or may
not take into account the curvature caused by the Earth's overall
spherical shape. Hence, there is no limitation as to the coordinate
system that may be used with the present invention, although it is
envisaged that Cartesian (x,y or x,y,z) coordinates or
latitude/longitude/altitude will be used most frequently because of
the way these inherently lend themselves to describing geographical
location, and because of the ease with which these coordinate
systems can be implemented digitally. Particularly representative
embodiments may utilise the WGS84 datum which is consistent with
the current GPS.
[0031] Those skilled in the art will know that GPS (global
positioning system) is the name of the satellite based navigation
system originally developed by the United States Department of
Defense. GPS is now used in a wide range of applications. A number
of systems also exist for increasing the accuracy of the location
readings obtained using GPS receivers. Some of these systems
operated by taking supplementary readings from additional
satellites and using these supplementary readings to "correct" the
original GPS location readings. These systems are commonly referred
to as "Satellite Based Augmentation Systems" (SBAS) and some
examples of SBASs are:
[0032] The United States'"Wide Area Augmentation System"
(WAAS),
[0033] The European Space Agency's "European Geostationary
Navigation Overlay Service" (EGNOS), and
[0034] The Japanese' "Multi-Functional Transportation Satellite"
(MFTS).
[0035] A number of "Ground Based Augmentation Systems" (GBASs) also
exist which help to increase the accuracy of GPS location readings
by taking additional readings from beacons located at known
locations on the ground. It will be understood that, throughout
this specification, all references to GPS include GPS when
augmented by supplementary systems such as SBASs, GBASs and the
like.
[0036] It is explained above that the controller (which controls
the vehicle) receives spatial data from the spatial database. In
this way, the data received by the controller from the database
forms at least part of the control inputs that the controller
operates on to control the vehicle (i.e. the spatial data forms at
least part of the inputs that drive the controller). The fact that
the controller operates directly on information that is inherently
associated with "real world" geographic location represents a
change in thinking compared with existing vehicle control systems.
In particular, it means that the control system of the present
invention "thinks" directly in terms of spatial location. Put
another way, in the control system of the present invention,
control parameters are defined in geographic space rather than the
space of an abstract vector. Consequently, the controller of the
present invention may be considered to be inherently
"multi-dimensional" or "spatial" in nature, as opposed to
"one-dimensional" or "linear" like the existing control systems
described in the background section above.
[0037] It is envisaged that at least some (and probably most) of
the components of the control system, including the controller,
will typically be implemented using commercially available
equipment and a generally conventional control architecture. For
instance, the controller may be implemented using equipment that
provides memory and a central processing unit to run the one or
more algorithms required to control the vehicle. Likewise, the
controller (and hence the control algorithm(s)) used in the present
invention may take any form suitable for controlling the steering
of a vehicle. Typically, closed loop or feedback type control will
be used at least in relation to some signal streams (i.e. in
relation to at least some of the vehicle variables being controlled
by the controller). However, open loop control may also be used, as
may feed-forward control structures wherein the spatial data
received by the controller from the spatial database is fed forward
to form part of the control outputs used to control the vehicle.
Where feedback type control is used, the control structure may
incorporate combinations of proportional, integral and differential
control, or a series of such (possibly nested) control loops.
However, no particular limitation is meant in this regard and the
person skilled in the art will appreciate that any form of suitable
control and/or controller may be used.
[0038] The control system may also incorporate conventional signal
processing and transmitting equipment, for example, for suitably
filtering incoming spatial data signals, and for transmitting
control signals from the controller to the vehicle's steering
system to steer the vehicle. The person skilled in the art will
appreciate that any suitable electric, mechanical, pneumatic or
hydraulic actuators, or combinations thereof, may be used with the
present invention. The actuators may be linked with the vehicle's
steering and drive systems to control the steering, acceleration,
deceleration etc of the vehicle in response to control signals
produced by the controller. Associated equipment such as amplifiers
and power sources may also be provided as required to amplify the
control signals, and to power the actuators. A wide range of power
sources may be used including batteries, generators, pumps etc
depending on the nature of the actuator(s) and the signals to be
amplified.
[0039] Whilst the present control system may operate using a
conventional form of controller and using at least some
commercially available equipment, the spatial database used to
store the spatial data and to provide the spatial data to the
controller may be different to other forms of databases used in
other areas. In other areas (including non-control related
applications such as those where data storage is the principal
objective), databases often contain the vast amounts of information
(in this case "information" is not used in its "spatial" sense) and
the information is generally stored in complex hierarchical
structures. Conceptually, these databases may be considered to be
"multi-levelled" in that an initial query may return only
relatively superficial level information, but this may in turn
allow the user to interrogate the database more deeply to obtain
more specific, linked or related information. This complex
structure means that these kinds of databases can take considerable
time (many seconds, minutes or even longer) to generate the
appropriate output in response to a query. Those skilled in the art
will appreciate that databases such as these, which take a
relatively long time to return information in response to a query,
may not be suitable for use in control systems such as the present
which require low latencies between variable inputs and control
outputs to thereby enable real-time control to be provided.
[0040] The spatial database used in the present invention will
suitably be adapted to provide the data to the controller at
control speed. In this sense, "control speed" means that the
database is able to provide the information at a rate of the same
order as the speed at which the controller repeats successive
cycles of the control algorithm (i.e. at a rate of the same order
as the "clock speed" of the controller). Ideally, the database will
be adapted to provide the data to the controller, and perhaps also
receive data from the controller and/or external sources, at every
successive cycle of the control algorithm (i.e. at the controller's
clock speed). However, in some embodiments it may be sufficient for
the database to be adapted to provide (and perhaps receive) data at
less than, but close to, the controller's clock speed (for example,
at every second or third successive cycle of the control
algorithm), provided that the rate is fast enough to provide the
controller with sufficiently up-to-date spatial information to
achieve adequate vehicle control performance. In cases where the
controller operates at different clock speeds for different data
signal streams, the database may be adapted to provide data at a
rate of the same order as one of those controller clock speeds. In
any event, the database should provide data to the controller at a
rate commensurate with the control loop bandwidth.
[0041] In practice, it is envisaged that the database may be
adapted to provide data to the controller at a rate of between 1 Hz
and 100 Hz. Given the speeds that vehicles such as agricultural
vehicles typically move at (generally less than 60 km/hr or 37.3
miles/hr), rates between 1 Hz and 20 Hz will almost always be
sufficient, and even rates between 3 Hz and 12 Hz may be sufficient
for vehicles moving at significantly less than 60 km/hr.
Nevertheless, those skilled in the art will recognise that the
necessary or achievable rates may vary depending on the level of
control precision and performance required in different
applications, the speed at which the vehicle in question moves, and
the capabilities of the available equipment used to implement the
control system.
[0042] Those skilled in the art will appreciate that because the
spatial database used in the present invention can provide spatial
data to the controller at control speed, and therefore forms part
of the system's overall configuration, the spatial database may be
considered to be "embedded" within the control system, rather than
external to it. This is particularly so in embodiments where
feedback type control is used, and the spatial database forms part
of the system's overall closed loop structure (i.e. in embodiments
where the spatial database forms part of the loop).
[0043] In order for the database to be able to provide (and, if
desired, also receive) data at the required rates, the form of the
database should allow the required rapid database access and
response times. Ideally, the database and all of the data that it
contains will be loaded into the control system's memory (i.e.
loaded into RAM). This way, the data will be directly accessible by
the controller's CPU (central processing unit), rather than
requiring a query to be sent to a remote disk or storage device
containing the data, the response to which would then need to be
loaded into RAM before being accessible by the CPU. However, it is
possible that the database could be located on a separate disk or
other storage device, particularly if the device is capable of
retrieving data in response to a query with sufficient speed such
as, for example, a disk device with RAM read/write cache.
[0044] It is envisaged that the amount of memory required to store
the spatial data relating to a particular field to be traversed by
the vehicle may be in the order of megabytes. By way of example
(given for illustrative purposes only), consider a straight wayline
that is 1 km long and which has 500 parallel swaths of
corresponding length. If the database is designed to incorporate
information pertaining to locations every 2 m along each of the 500
swaths, this corresponds to 501.times.500=250,500 locations. When
the data is structured within the database in the manner described
further below, this may correspond to approximately 4 MB of memory
required to store the coordinates of each point. However, it is
also envisaged that as the nature and complexity of the data
required to be stored in the database increases, the required
amount of memory may increase to hundreds of megabytes or
gigabytes. Devices which provide this amount of memory are (or are
at least becoming) commercially available.
[0045] The speed of the database may be assisted by the way in
which the data is arranged (i.e. stored) within the database. A
wide range of methods and algorithms are known for arranging data
(i.e. for assigning appropriate "indices" and the corresponding
memory allocations to individual items of data) within databases,
and the particular method chosen depends on the nature of the data,
and the way and speed with which the database is to respond to a
query. For the complex hierarchical "multi-levelled" databases
described above, the data should be arranged so as to enable the
database to collate and deliver all relevant information relating
to a complex query. However, as explained above, the requirement
for those databases to be able to process complex queries leads to
potentially long lag times which may be undesirable in the context
of vehicle control applications. Therefore, the spatial database
used in the present invention can store data in a "single-level" or
"flat" structure according to the geographical location that
particular items of data relate to.
[0046] Some algorithms which could be used to arrange the spatial
data within the database include the algorithms commonly referred
to by the names "Grid-indexing", "Quadtree" or "R-tree". However,
in other embodiments of the invention data may be arranged within
the database using a form of algorithm that will be referred to as
a "spatial hash-key" algorithm. A spatial hash-key algorithm maps
physical locations (based on their "real world" coordinates) into
one-dimensional "hash-keys". The "hash-key" for each location is a
string of characters that can be stored in the database's hash
table and retrieved in response to a query.
[0047] Properties of the spatial hash-key algorithm may
include:
[0048] points which are close to each other in the real world
should have closely related hash keys (i.e. the algorithm should
maintain "locality"), the algorithm should operate using whatever
coordinate system the control system uses to represent the
region,
[0049] the algorithm should be adapted for digital implementation
(hence, it should be adapted to operate using integer or
floating-point numbers, preferably with 64-bit "double" precision
or better)
[0050] the algorithm should be fast to compute.
[0051] It is explained above that the control system of the present
invention, and ideally the spatial database, may be adapted to
receive updated data from the controller and/or an external source.
The spatial database can be adapted to receive the updated
information at control speed. Data received from the controller may
include or may be used to generate, for example, estimates of the
vehicle's predicted state (i.e. its speed, position, orientation
etc) at an upcoming location based on its current instantaneous
state at a particular location. The external sources may include
GPS, INS, or any other inertial, visual or other system used for
obtaining information relating to the state of the vehicle or other
aspects of the region (such as obstacles close to the vehicle).
Data received in this way may be (at least initially) recorded in
its unprocessed or "raw" form in the database. This unprocessed
data may be fed directly back into the controller, or the
respective streams of incoming data (possibly relating to disparate
variables) may be filtered using a Kalman filter or some other
similar digital signal processing technique to obtain a
statistically optimised estimate of the state of the vehicle and
its proximate surroundings as it travels. This optimised estimate
of the vehicle's state at a particular location may then be fed
into the controller. The use of statistically optimised estimates
and data may help to improve control performance.
[0052] According to a further broad form, the invention resides in
a closed loop vehicle control system comprising
[0053] a spatial database,
[0054] a controller adapted to receive spatial data from the
spatial database at control speed, the controller controlling the
steering of the vehicle,
[0055] wherein updated spatial data is fed back into the control
system.
[0056] In yet another broad form, the invention resides in a method
for controlling a vehicle comprising
[0057] entering spatial data relating to a region to be traversed
by the vehicle into a spatial database,
[0058] providing spatial data from the spatial database to a
controller at control speed to control the vehicle as the vehicle
traverses the region, and
[0059] entering updated spatial data into the spatial database as
the vehicle traverses the region.
[0060] In yet a further broad form, the invention resides in a
vehicle control system comprising
[0061] a spatial database,
[0062] a controller adapted to receive spatial data from the
spatial database,
[0063] the controller using the spatial data from the spatial
database to control the steering of the vehicle.
[0064] It will be appreciated that all preferred features and
aspects of the invention described with particular reference to one
or other broad form of the invention, may also apply equally to all
other forms of the invention, unless the context dictates
otherwise.
BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Certain embodiments, aspects and features of the invention
will now be described and explained by way of example and with
reference to the drawings. However, it will be clearly appreciated
that these descriptions and examples are provided to assist in
understanding the invention only, and the invention is not limited
to or by any of the embodiments, aspects or features described or
exemplified.
[0066] FIG. 1 schematically represents the difference between the
vehicle's actual spatial location and what is "seen" by existing
forms of "one-dimensional" controllers such as those described in
the background section above,
[0067] FIG. 2 is a pictorial representation of an agricultural
vehicle having a control system in accordance with one particular
embodiment of the present invention,
[0068] FIG. 3 illustrates the physical meaning of certain
parameters controlled by some versions of the present control
system, namely the "cross-track error", the "heading error" and the
"curvature error",
[0069] FIG. 4 is a schematic "block-diagram" representation of an
overall control system structure that may be used in representative
embodiments of the present invention,
[0070] FIG. 5 is a schematic representation of the "controller"
block that may be used in representative embodiments such as that
shown in FIG. 4,
[0071] FIG. 6 is a further schematic "block-diagram" representation
of an overall control system structure that may be used with
alternative representative embodiments of the invention which
incorporate additional features not shown in FIG. 4.
[0072] FIG. 7 is a block diagram representation of the state space
representation used in the digital implementation of certain
aspects of the control system.
[0073] FIG. 8 shows an example trajectory of an agricultural
vehicle, and the coordinates corresponding to different points
along the trajectory using a simplified integer based coordinate
system.
[0074] FIG. 9 shows a similar example trajectory of an agricultural
vehicle to that shown in FIG. 8, except that the coordinate system
is similar in format to the WGS84 coordinate used by current
GPS.
[0075] FIG. 10 illustrates the way in which numbers are represented
in the IEEE 754 standard double-precision floating-point format,
and
[0076] FIG. 11 is a "flow-diagram" illustrating the way a
particularly preferred spatial hash algorithm may be used to
generate hash keys for the coordinates in FIG. 9.
BEST MODE
[0077] As described in the background section above, one of the
problems with existing vehicle control systems is that they are
inherently "one-dimensional" or "linear" in nature. The inherent
"linear" nature of existing control systems is illustrated
schematically in FIG. 1. Whilst the "real world" spatial geometry
of the respective swaths shown on the left in FIG. 1 may have been
calculated, nevertheless from the control system's point of view at
any given time the controller only "knows" that the vehicle is on
the nth swath and that it has been moving along that swath for a
known amount of time with known speed. Hence, at a fundamental
level, the controller does not inherently know where the vehicle is
located in space. This is represented graphically in FIG. 1.
[0078] Next, FIG. 2 shows an agricultural vehicle 1 having a
control system in accordance with one embodiment of the present
invention. In FIG. 2, the agricultural vehicle 1 is a tractor
towing an implement 2. The implement 2 could be a plough,
harvester, seed sower, leveller, agricultural chemical
applicator/dispenser or any other kind of agricultural implement.
Furthermore, the embodiment of the invention shown in FIG. 2 could
equally be applied on other kinds of vehicles operating in other
areas, for example cars, mine-trucks, airport tarmac vehicles,
etc.
[0079] The components of the control system in the particular
embodiment shown in FIG. 2 include a main control unit 3, a GPS
antenna 4 and actuators 5. The main control unit 3 houses the
spatial database and also the electronic hardware used to implement
the controller. The main control unit 3 may be an industrial
computer (for example an industrial PC) capable of running other
applications in addition to the vehicle control system.
Alternatively, the main control unit 3 may be a purpose-built unit
containing only the hardware required to run the controller, the
spatial database and the other components of the vehicle control
system.
[0080] The main control unit 3 receives GPS signals from the GPS
antenna 4, and it uses these (typically in combination with
feedback and/or other external spatial data signals) to generate a
control signal for steering the vehicle. The control signal will
typically be made up of a number of components or streams of data
relating to the different parameters of the vehicle being
controlled, for example the vehicle's "cross-track error", "heading
error", "curvature error", etc. These parameters will be described
further below. The control signal is amplified using suitable
signal amplifiers (not shown) to create a signal that is
sufficiently strong to drive the actuators 5. The actuators 5 are
interconnected with the vehicle's steering mechanism (not shown)
such that the actuators operate to steer the vehicle as directed by
the control signal.
[0081] In some embodiments, further actuators (not shown) may also
be provided which are interconnected with the vehicle's accelerator
and/or braking mechanisms, and the control signal may incorporate
components or signal streams relating to the vehicle's forward
progress (i.e. its forward speed, acceleration, deceleration etc).
In these embodiments, the component(s) of the control signal
relating to the vehicle's forward progress may also be amplified by
amplifiers (not shown) sufficiently to cause the actuators which
are interconnected with the accelerator/braking mechanism to
control the vehicle's acceleration/deceleration in response to the
control signal.
[0082] The vehicle 1 may also be optionally provided with one or
more visual sensors 6, one or more inertial sensors 7 and a user
terminal 8. One form of visual sensor 6 that may be used may
operate by receiving images of the ground beneath the vehicle,
preferably in rapid succession, and correlating the data pertaining
to respective successive images to obtain information relating to
the vehicle's motion. Other forms of visual sensor may also be used
including LIDAR (Light Detection and Ranging) or sensors which
operate using machine vision and/or image analysis. If present, the
one or more inertial sensors 7 will typically include at least one
gyroscope (e.g. a rate gyroscope), although the inertial sensors 7
could also comprise a number of sensors and components (such as
accelerometers, tilt sensors and the like) which together form a
sophisticated inertial navigation system (INS). The vehicle may be
further provided with additional sensors (not shown) such as
sensors which receive information regarding the location of the
vehicle relative to a fixed point of known location in or near the
field, magnetometers, ultrasonic range and direction finding and
the like. The data generated by these additional sensors may be fed
into the database and used by the control system to control the
vehicle as described below.
[0083] In embodiments where the main control unit 3 comprises an
industrial PC or the like, the user terminal 8 may comprise a full
computer keyboard and separate screen to enable the user to utilise
the full functionality of the computer. However, in embodiments
where the main control unit is a purpose-built unit containing only
hardware relating to the vehicle's control system, the terminal 8
may comprise, for example, a single combined unit having a display
and such controls as may be necessary for the user to operate the
vehicle's control system. Any kind of controls known by those
skilled in this area to be suitable may be used on the main control
unit, including keypads, joysticks, touch screens and the like.
[0084] In FIG. 2, the user terminal 8 is positioned in the vehicle
cabin so that it can be operated by the driver as the vehicle
moves. However, those skilled in the art will recognise that the
present control system could also be operated by wireless remote
control, meaning that the user terminal 8 could alternatively be
totally separate from the vehicle and could operate the vehicle's
control system from a remote location. It is also envisaged that a
single remote user terminal 8 may be used to wirelessly interface
with the control systems of multiple vehicles (possibly
simultaneously) so that the user can control multiple moving
vehicles from the one remote terminal.
[0085] In order to control the steering of the vehicle, there are
three parameters that should be controlled. These are the
"cross-track error", the "heading error" and the "curvature error".
The physical meaning of these parameters can be understood with
reference to FIG. 3. The "cross-track error" is the lateral
difference between the vehicle's actual position, and its desired
position. This is illustrated by the "{" bracket in FIG. 3. The
"heading error" is the difference between the vehicle's actual
instantaneous direction of motion h (i.e. its actual compass
heading), and its desired instantaneous direction of motion H. The
heading error is given by:
Heading Error=H-h
[0086] Those skilled in the art will recognise that both h and H
are inherently directional quantities.
[0087] Finally, the "curvature error" is the difference between the
actual instantaneous radius of curvature r of the vehicle's motion
and the desired instantaneous radius of curvature R. The curvature
error is given by:
Curvature Error=1/R-1/r
[0088] It will also be clearly appreciated that there may be many
other vehicle variables or parameters which also need to be
controlled if, for example, acceleration/deceleration or the
vehicle's mode of equipment operation are also to be
controlled.
[0089] Referring next to FIG. 4, it can be seen that a vehicle
control system in accordance with one particular embodiment of the
invention comprises:
[0090] a task path generator,
[0091] a spatial database,
[0092] at least one external spatial data source,
[0093] a vehicle attitude compensation module,
[0094] a position error generator,
[0095] a controller, and
[0096] actuators to control (steer) the vehicle.
[0097] In the overall operation of the control system, the desired
path trajectory for the vehicle is first entered into the control
system by the user via the user terminal 8. The task path generator
then interprets this user-defined path definition and converts it
into a series of points of sufficient spatial density to adequately
represent the desired path to the requisite level of precision. The
task path generator typically also defines the vehicle's desired
trajectory along the user-defined path, for example, by generating
a desired vehicle position, a desired heading H and a desired
instantaneous radius of curvature R for each point on the path.
This information is then loaded into the spatial database. The way
in which this and other spatial information is stored within the
database in representative embodiments, and in particular the way
in which pieces of data are given memory allocations according to
their spatial location, is described further below.
[0098] As the vehicle moves along the user-defined path, it will
invariably experience various perturbations in its position and
orientation due to, for example, bumps, potholes, subsidence
beneath the vehicle's wheels, vehicle wheel-spin, over/under-steer
etc. Those skilled in this area will recognise that a huge range of
other similar factors can also influence the instantaneous position
and orientation of the vehicle as it moves. One of the purposes of
the present control system is to automatically correct for these
perturbations in position and orientation to maintain the vehicle
on the desired path (or as close to it as possible).
[0099] As the vehicle moves, the control system progressively
receives updated information regarding spatial location from the
external spatial data sources. The external spatial data sources
will typically include GPS. However, a range of other spatial data
sources may also be used in addition to, or in substitute for GPS.
For example, the inertial navigation systems (INS), visual
navigation systems etc described above may also be used as external
data sources in the present control system.
[0100] Those skilled in the art will recognise that the spatial
data collected by the external spatial data sources actually
pertains to the specific location of the external spatial data
receivers, not necessarily the vehicle/implement reference location
itself (which is what is controlled by the control system). In FIG.
2, the reference location is on the vehicle 1 and is indicated by
the intersection (i.e. the origin) of the roll, pitch and yaw axes.
In other embodiments, the reference location may be located
elsewhere on the vehicle, or on the implement 2 etc. In any event,
to illustrate this point, it will be seen that the GPS antenna 4 in
FIG. 2 is located on the roof of the vehicle some distance from the
vehicle's reference point. Therefore, the spatial data collected by
the GPS antenna actually relates to the instantaneous location of
the vehicle's roof, not the location of the vehicle's reference
point. Likewise, the spatial data collected by the visual sensor 6
actually pertains to the particular location of the visual sensor
(slightly out in front of the vehicle in FIG. 2).
[0101] In addition to this, changes in the vehicle's attitude will
also influence the spatial position readings received by the
different receivers. For example, if one of the vehicle's wheels
passes over, or is pushed sideways by a bump, this may cause the
vehicle to rotate about at least one (and possibly two or three) of
the axes shown in FIG. 2. This will in turn change the relative
position of the spatial data receiver(s) such as GPS antenna 4 with
respect to the reference location on the vehicle or implement. This
can be used (typically in combination with other sources of
external spatial data or "feedback" data) to determine the
orientation of the vehicle. The orientation of the vehicle may be
considered to be the relative orientation of the vehicle's axes in
space.
[0102] In order to compensate for the difference in position
between the vehicle's reference point and the location of the
spatial data receiver(s), and also to account for changes in the
vehicle's orientation, a vehicle attitude compensation module is
provided. This is shown in FIG. 4. The vehicle attitude
compensation module converts all readings taken by the various
spatial data receivers (which relate to the different specific
locations of the receivers) into readings pertaining to the spatial
location and orientation of the vehicle's reference point. This
data pertaining to the spatial location and orientation of the
vehicle's reference point is then fed into the spatial
database.
[0103] Those skilled in the art will recognise that the one or more
external spatial data sources will progressively receive updated
data readings in rapid succession (e.g. in "real time" or as close
as possible to it). These readings are then converted by the
vehicle attitude compensation module and fed into the spatial
database. The readings may also be filtered as described above.
Therefore, whilst each reading from each spatial data source is
received, converted (ideally filtered) and entered into the spatial
database individually, nevertheless the rapid successive way in
which these readings (possibly from multiple "parallel" data
sources) are received, converted and entered effectively creates a
"stream" of incoming spatial data pertaining to the vehicle's
continuously changing instantaneous location and orientation. In
order to provide sufficient bandwidth, successive readings from
each external spatial data source should be received and converted
with a frequency of the same order as the clock speed (or at least
one of the clock speeds) of the controller, typically 3 Hz-12 Hz or
higher.
[0104] Referring again to FIG. 4, the position error generator next
receives information from the spatial database. The information it
receives from the database includes:
[0105] the vehicle's desired position, heading H and instantaneous
radius of curvature R. It will be recalled that this information is
originally generated by the task path generator and then entered
into the spatial database, based on the user-defined path
trajectory.
[0106] And the vehicle's actual position, heading h and
instantaneous radius of curvature r.
[0107] This information is based on spatial data progressively
received from the external spatial data sources as described above,
and typically also on data received through feedback.
[0108] The position error generator then uses this information to
calculate an instantaneous "error term" for the vehicle. The "error
term" incorporates the vehicle's instantaneous cross-track error,
heading error and curvature error (as described above). The error
term is then fed into the controller. The controller is shown in
greater detail in FIG. 5.
[0109] From FIG. 5 it can be seen that the controller incorporates
a cross-track error PID controller, a heading error PID controller
and a curvature error PID controller. The PID controllers used with
the present invention are of a conventional form that will be well
understood by those skilled in this area and need not be described
in detail. The output from the cross-track error, heading error and
curvature error PID controllers then passes through a curvature
demand signal integrator. The output from the PID controllers is
therefore integrated in order to generate a curvature demand
signal. This curvature demand signal is thus the "control signal"
which is amplified by amplifiers (not shown) before proceeding to
drive the actuators as required. In other words, the signal
obtained by integrating the output from the PID controllers is
amplified and sent to the actuators in the form of a curvature
demand to change the vehicle's steering angle and hence steer the
vehicle back onto the desired path. Finally, the change in vehicle
pose etc caused by the control driven change in steering angle is
registered via the updated information received through the
external data sources (GPS etc) and the vehicle's new position,
heading and instantaneous radius of curvature are re-entered into
the spatial database to complete control system's overall closed
loop control structure. It will be noted that the arrows extending
from the actuators/steering mechanism to the external data sources
in FIG. 4 are dashed rather than solid lines. This is to indicate
that, whilst there is no actual control signal or other data flow
from the actuators/steering mechanism to the external data sources,
there is nevertheless a causal link between the change in vehicle
pose etc caused by the control driven change in steering angle and
the updated information received through the external data
sources.
[0110] In FIG. 6, there is shown a slightly more elaborate
embodiment of the control system.
[0111] The embodiment shown in FIG. 6 is generally the same as that
shown in FIG. 4, except that the embodiment in FIG. 6 incorporates
an optimising filter and an external obstacle detection input. The
optimising filter can operate to statistically optimise at least
some of the spatial data contained in the spatial data base. Also,
the filter will generally operate as an "observer", meaning that it
does not form part of the control loop. Rather, the filter will
typically reside outside the control loop and it will generally
operate by taking data directly from the database and returning
optimise data directly into the database, as shown in FIG. 6. More
specifically, the filter will take the updated "feedback" data that
re-enters the database from the control loop (described above)
together with the updated spatial data obtained from the external
spatial data sources (after it has been processed by the vehicle
attitude compensation module) and it will then use these disparate
streams of data to calculate a statistically optimised updated
estimate of, for example, the vehicle's instantaneous position,
heading and radius of curvature. The filter will typically comprise
a Kalman filter.
[0112] The external obstacle detection input may comprise any form
of vision based, sound based or other obstacle detection means, and
the obstacle detection data may be converted by the vehicle
attitude compensation module (just like the other sources of
external data discussed above) and then fed into the spatial
database. Where the control system incorporates obstacle detection,
it is then necessary for the task path generator to be able to
receive updated information from the spatial database. This is so
that if an obstacle is detected on the desired path, an alternative
path that avoids the obstacle can be calculated by the task path
generator and re-entered into the database. The ability of the task
path generator to also receive data from the spatial database is
indicated by the additional arrow from the spatial database to the
task path generator in FIG. 6.
[0113] FIGS. 4-6 graphically represent the operation of the control
system. However, it is also useful to consider the way in which the
vehicle's parameters and dynamics are represented for the purposes
of implementing the control system. Those skilled in the art will
recognise that a range of methods may be used for this purpose.
However, it is considered that one method is to represent the
parameters and dynamics in "state space" form.
[0114] In state space representations, the variables or parameters
used to mathematically model the motion of the vehicle, or aspects
of its operation, are referred to as "states" x.sub.i. In the
present case, the states may include the vehicle's position (x,y),
velocity
( x t , y t ) ##EQU00001##
heading h, radius of curvature r etc. Hence the states may include
x.sub.i=x, x.sub.2=y, x.sub.3=h, x.sub.4=h,
x 5 = x t = dx 1 t , x 6 = y t = x 2 t ##EQU00002##
Etc. However, it will be appreciated that the choice of states is
never unique, and the meaning and implications of this will be well
understood by those skilled in the art.
[0115] The values for the individual states at a given time are
represented as the individual entries in an n.times.1 "state
vector":
X(t)=[x.sub.1(t)x.sub.2(t)x.sub.3(t)x.sub.4(t) . . .
x.sub.n(t)].sup.T
where n is the number of states.
[0116] In general, the mathematical model used to model the
vehicle's motion and aspects of its operation will comprise a
series of differential equations. The number of equations will be
the same as the number of states. In some cases, the differential
equations will be linear in terms of the states, whereas in other
situations the equations may be nonlinear in which case they must
generally be "linearised" about a point in the "state space".
Linearisation techniques that may be used to do this will be well
known to those skilled in this area.
[0117] Next, by noting that any j.sup.th order linear differential
equations can be re-written equivalently as a set j first order
linear differential equations, the linear (or linearised) equations
that represent the model can be expressed using the following
"state" equation:
t ( X _ ( t ) ) = A X _ ( t ) + B U _ ( t ) + E w _ ( t )
##EQU00003##
where:
[0118] A is an n.times.n matrix linking the state time derivatives
to the states themselves,
[0119] U(t) is an m.times.1 matrix containing the external
"forcing" inputs in the mathematical model,
[0120] B is an n.times.m matrix linking the state derivatives to
the inputs,
[0121] m is the number of inputs,
[0122] Ew(t) is a quantity (represented by an n.times.1 vector)
called the "process noise".
[0123] The process noise represents errors in the model and vehicle
dynamics which exist in the actual vehicle but which are not
accounted for in the model. As Ew(t) represents an unknown
quantity, its contents are not known. However, for reasons that
will be understood by those skilled in this area, in order to allow
statistically optimised signal processing and state estimation
Ew(t) is generally assumed to be Gaussian, white, have zero mean
and to act directly on the state derivatives. It is also assumed
that the process noise element associated with each individual
state is uncorrelated with the process noise element of the other
states.
[0124] The quantities that are desired to be known about the
vehicle (the real values for which are generally also measured from
the vehicle itself, if possible) are the outputs y.sub.1 from the
model. Each of the outputs generated by the linear (or linearised)
model comprises a linear combination of the states x, and inputs
u.sub.i, and so the outputs can be defined by the "output" or
"measurement" equation:
Y(t)=CX(t)+DU(t)Mv(t)
[0125] Where C is a j.times.n matrix linking the outputs to the
states,
[0126] D is a j.times.m matrix linking the outputs to the
inputs,
[0127] j is the number of outputs, and
[0128] Mv(t) is a quantity (represented by an n.times.1 vector)
called the "measurement noise".
[0129] The measurement noise represents errors and noise that
invariably exist in measurements taken from the actual vehicle.
Like Ew(t) above, Mv(t) is assumed to be Gaussian, white, have zero
mean, to act directly on the state derivatives and to be
uncorrelated with the process noise or itself.
[0130] Next, it will be noted that both the state equation and the
measurement equation defined above are continuous functions of
time. However, continuous time functions do not often lend
themselves to easy digital implementation (such as will generally
be required in implementing the present invention) because digital
control systems generally operate as recursively repeating
algorithms. Therefore, for the purpose of implementing the
equations digitally, the continuous time equations may be converted
into the following recursive discrete time equations by making the
substitutions set out below and noting that (according to the
principle of superposition) the overall response of a linear system
is the sum of the free (unforced) response of that system and the
responses of that system due to forcing/driving inputs. The
recursive discrete time equations are:
X.sub.k+1=FX.sub.k+GU.sub.k+1+Lw.sub.k+1
Y.sub.k+1=ZX.sub.k+JU.sub.k+1+Nv.sub.k+1
[0131] where k+1 is the time step occurring immediately after time
step k,
[0132] Z=C, J=D and Nv is the discrete time analog of the
continuous time measurement noise Mv(t).
[0133] F is a transition matrix which governs the free response of
the system. F is given by:
F=e.sup.A.delta.
[0134] GU.sub.k+1 is the forced response of the system, i.e. the
system's response due to the driving inputs. It is defined by the
convolution integral as follows:
GU.sub.k+1=.intg..sub.0.sup..DELTA.te.sup.A(.DELTA.t-.tau.)dt
[0135] Similarly, the quantity Lw.sub.k+1 is the (forced) response
of the system due to the random "error" inputs that make up the
process noise. Hence, conceptually this quantity may be defined
as:
Lw.sub.k+1=.intg..sub.0.sup..DELTA.te.sup.A(.DELTA.t-.tau.)dtE.sub.w(t.s-
ub.k+1+.tau.)dt
[0136] However, as noted above, the quantity Ew(t) is not
deterministic and so the integral defining Lw.sub.k+1 cannot be
performed (even numerically). It is for this reason that it is
preferable to use statistical filtering techniqu.tau.es such as a
"Kalman Filter" to statistically optimise the states estimated by
the mathematical model.
[0137] In general, a "Kalman Filter" operates as a
"predictor-corrector" algorithm. Hence, the algorithm operates by
first using the mathematical model to "predict" the value of each
of the states at time step k+1 based on the known inputs at time
step k+1 and the known value of the states from the previous time
step k. It then "corrects" the predicted value using actual
measurements taken from the vehicle at time step k+1 and the
optimised statistical properties of the model. In summary, the
Kalman Filter comprises the following equations each of which is
computed in the following order for each time step:
X _ k + 1 k = F X _ k k + G U _ k + 1 P k + 1 k = FP k k F T + Q K
k + 1 = P k + 1 k Z T ( ZP k + 1 k Z T + R ) - 1 Y _ k + 1 = Z X _
k + 1 k + J U _ k + 1 v _ k + 1 = Y _ ^ k + 1 - Y _ k + 1 }
predictor X _ k + 1 k + 1 = X _ k + 1 k + K k + 1 v k + 1 P k + 1 k
+ 1 = ( I - K k + 1 Z ) P k + 1 k } corrector ##EQU00004##
[0138] where the notation k+1|k means the value of the quantity in
question at time step k+1 given information from time step k.
Similarly, k+1|k+1 means the value of the quantity at time step k+1
given updated information from time step k+1. [0135]P is the
co-variance in the difference between the estimated and actual
value of X. [0136]Q is the co-variance in the process noise.
[0137]K is the "Kalman gain" which is a matrix of computed
coefficients used to optimally "correct" the initial state
estimate. [0138]R is the co-variance in the measurement noise.
[0139] is a vector containing measurement values taken from the
actual vehicle.
[0139] is a quantity called the "innovation" which is the
difference between the measured values actually taken from the
vehicle and values for the corresponding quantities estimated by
the model.
[0140] The operation of the discrete time state space equations
outlined above, including the Kalman gain and the overall feedback
closed loop control structure, are represented graphically in FIG.
7.
[0141] In relation to the spatial database, it is mentioned above
that a wide range of methods are known for arranging data within
databases. One commonly used technique is to provide a "hash
table". The hash table typically operates as a form of index
allowing the computer (in this case the control system CPU) to
"look up" a particular piece of data in the database (i.e. to look
up the location of that piece of data in memory). In the context of
the present invention, pieces of data pertaining to particular
locations along the vehicle's path are assigned different hash keys
based on the spatial location to which they relate. The hash table
then lists a corresponding memory location for each hash key. Thus,
the CPU is able to "look up" data pertaining to a particular
location by looking up the hash key for that location in the hash
table which then gives the corresponding location for the
particular piece of data in memory. In order to increase the speed
with which these queries can be carried out, the hash keys for
different pieces of spatial data can be assigned in such a way that
"locality" is maintained. In other words, points which are close to
each other in the real world should be given closely related
indices in the hash table (i.e. closely related hash keys).
[0142] The spatial hash algorithm used to generate hash keys for
different spatial locations in representative embodiments of the
present invention may be most easily explained by way of a series
of examples. To begin, it is useful to consider the hypothetical
vehicle path trajectory shown in FIG. 8. In FIG. 8, the successive
points which define the path are described by a simplified integer
based (X,Y) coordinate system. Hence, in FIG. 8, the vehicle moves
in the X direction along the entire length of the first swath from
(0,0) to (4,0), before moving up in the Y direction to then move
back along the second swath in the opposite direction from (4,1) to
(0,1), etc.
[0143] As outlined above, in the present invention all data is
stored within the spatial database with reference to spatial
location. Therefore, it is necessary to assign indices or "hash
keys" to each piece of data based on the spatial location to which
each said piece of data relates. However, it will be recalled that
the hash table must operate by listing the hash key for each
particular spatial location together with the corresponding memory
location for data pertaining to that spatial location. Therefore,
the hash table is inherently one-dimensional, and yet it must be
used to link hash keys to corresponding memory allocations for data
that inherently pertains to two-dimensional space.
[0144] One simple way of overcoming this problem would be to simply
assign hash keys to each spatial location based only on, say, the Y
coordinate at each location. The hash keys generated in this way
for each point on the vehicle path in FIG. 8 are given in Table 1
below.
TABLE-US-00001 TABLE 1 Spatial Hash Key Generated Using Only the Y
Coordinate (X, Y) Hash key (X, Y) Hash key coor- (hexa- Hash key
coor- (hexa- Hash key dinates decimal) (decimal) dinates decimal)
(decimal) (0, 0) 0 .times. 0 0 (3, 2) 0 .times. 2 2 (1, 0) 0
.times. 0 0 (4, 2) 0 .times. 2 2 (2, 0) 0 .times. 0 0 (0, 3) 0
.times. 3 3 (3, 0) 0 .times. 0 0 (1, 3) 0 .times. 3 3 (4, 0) 0
.times. 0 0 (2, 3) 0 .times. 3 3 (0, 1) 0 .times. 1 1 (3, 3) 0
.times. 3 3 (1, 1) 0 .times. 1 1 (4, 3) 0 .times. 3 3 (2, 1) 0
.times. 1 1 (0, 4) 0 .times. 4 4 (3, 1) 0 .times. 1 1 (1, 4) 0
.times. 4 4 (4, 1) 0 .times. 1 1 (2, 4) 0 .times. 4 4 (0, 2) 0
.times. 2 2 (3, 4) 0 .times. 4 4 (1, 2) 0 .times. 2 2 (4, 4) 0
.times. 4 4 (2, 2) 0 .times. 2 2
[0145] The prefix "0x" indicates that the numbers in question are
expressed in hexadecimal format. This is a conventional
notation.
[0146] Those skilled in the art will recognise that the above
method for generating hash keys is far from optimal because there
are five distinct spatial locations assigned to each different hash
key. Furthermore, in many instances, this method assigns the same
hash key to spatial locations which are physically remote from each
other. For instance, the point (0,1) is distant from the point
(4,1), and yet both locations are assigned the same hash key. An
identically ineffective result would be obtained by generating a
hash key based on only the X coordinate.
[0147] An alternative method would be to generate hash keys by
concatenating the X and Y coordinates for each location. The hash
keys generated using this method for each point on the vehicle path
in FIG. 8 are given in Table 2 below.
TABLE-US-00002 TABLE 2 Hash Keys Generated by Concatenating the X
and Y Coordinates (X, Y) Hash key (X, Y) Hash key coor- (hexa- Hash
key coor- (hexa- Hash key dinates decimal) (decimal) dinates
decimal) (decimal) (0, 0) 0 .times. 0 0 (3, 2) 0 .times. 302 770
(1, 0) 0 .times. 100 256 (4, 2) 0 .times. 402 1026 (2, 0) 0 .times.
200 512 (0, 3) 0 .times. 3 3
[0148] In order to understand how the numbers listed in Table 2
above were arrived at, it is necessary to recognise that in the
digital implementation of the present control system, all
coordinates will be represented in binary. For the purposes of the
present example which relates to the simplified integer based
coordinate system in FIG. 8, a simplified 8-bit binary
representation has been used.
[0149] Hence, to illustrate the operation of the spatial hash key
algorithm used to generate the numbers in Table 2, consider the
point (3,3). Those skilled in the art will understand that the
decimal number 3 may be written as 11 in binary notation.
Therefore, the location (3,3) may be rewritten in 8-bit binary
array notation as (00000011,00000011). Concatenating these binary
coordinates then gives the single 16-bit binary hash key
0000001100000011 which can equivalently be written as the
hexadecimal number 0x303 or the decimal number 771. The process of
converting between decimal, binary and hexadecimal representations
should be well known to those skilled in the art and need not be
explained.
[0150] It will be noted from Table 2 above that concatenating the X
and Y coordinates leads to unique hash keys (in this example) for
each spatial location. However, the hash keys generated in this way
are still somewhat sub-optimal because points which are located
close to each other are often assigned vastly differing hash keys.
For example, consider the points (0,0) and (1,0). These are
adjacent point in the "real world". However, the hash keys assigned
to these points using this method (written in decimal notation) are
0 and 256 respectively. In contrast, the point (0,4) is much
further away from (0,0) and yet it is assigned the much closer hash
key 4. Therefore, this algorithm does not maintain "locality", and
an alternative algorithm would be preferable.
[0151] Yet a further method for generating hash keys is to use a
technique which shall hereinafter be referred to as "bitwise
interleaving". As for the previous example, the first step in this
technique is to represent the (X,Y) coordinates in binary form.
Hence, using the 8-bit binary array representation discussed above,
the point (X,Y) may be re-written in 8-bit binary array notation as
(X.sub.1X.sub.2X.sub.3X.sub.4X.sub.5X.sub.6X.sub.7X.sub.8,
Y.sub.1Y.sub.2Y.sub.3Y.sub.4Y.sub.5Y.sub.6Y.sub.7Y.sub.8). Next,
rather than concatenating the X and Y coordinates to arrive at a
single 16-bit binary hash key, the successive bits from the X and Y
binary coordinates are alternatingly "interleaved" to give the
following 16-bit binary hash key
X.sub.1Y.sub.1X.sub.2Y.sub.2X.sub.3Y.sub.3X.sub.4Y.sub.4X.sub.5Y.sub.-
5X.sub.6Y.sub.6X.sub.7YX.sub.8Y.sub.8. The hash keys generated
using this method for each point on the vehicle path in FIG. 8 are
given in Table 3 below.
TABLE-US-00003 TABLE 3 Hash Keys Generated by "Bitwise
Interleaving" the X and Y Coordinates (X, Y) (X, Y) Hash key (X, Y)
Hash key coor- (hexa- Hash key coor- (hexa- Hash key dinates
decimal) (decimal) dinates decimal) (decimal) (0, 0) 0 .times. 0 0
(3, 2) 0 .times. e 14 (1, 0) 0 .times. 2 2 (4, 2) 0 .times. 24 36
(2, 0) 0 .times. 8 8 (0, 3) 0 .times. 5 5 (3, 0) 0 .times. a 10 (1,
3) 0 .times. 6 7 (4, 0) 0 .times. 20 32 (2, 3) 0 .times. d 13 (0,
1) 0 .times. 1 1 (3, 3) 0 .times. f 15 (1, 1) 0 .times. 3 3 (4, 3)
0 .times. 25 37 (2, 1) 0 .times. 9 9 (0, 4) 0 .times. 10 16 (3, 1)
0 .times. b 11 (1, 4) 0 .times. 12 18 (4, 1) 0 .times. 21 33 (2, 4)
0 .times. 18 24 (0, 2) 0 .times. 4 4 (3, 4) 0 .times. 1a 26 (1, 2)
0 .times. 6 6 (4, 4) 0 .times. 30 48 (2, 2) 0 .times. e 12
[0152] To further illustrate the operation of the spatial hash
algorithm used to generate the numbers in Table 3, consider the
point (3,4). As noted above, the decimal number 3 may be written as
11 in binary notation. Similarly, decimal number 4 is written as
100 in binary. Therefore, the location (3,4) may be rewritten in
8-bit binary array notation as (00000011,00000100). Bitwise
interleaving these binary coordinates then gives the single 16-bit
binary hash key 0000000000011010, which can equivalently be written
as the hexadecimal number 0x1a or the decimal number 26.
[0153] From Table 3 it will be seen that generating hash keys by
"bitwise interleaving" the X and Y coordinates leads to unique hash
keys (in this example) for each spatial location. Also, the hash
keys generated in this way satisfy the requirement that points
which are close together in the real world are assigned closely
related hash keys. For example, consider again the points (0,0) and
(1,0). The hash keys now assigned to these points by "bitwise
interleaving" (when written in decimal notation) are 0 and 2
respectively. Furthermore, the point (0,1) which is also nearby is
also assigned the closely related hash key 1. Conversely, points
which are separated by a considerable distance in the real world
are given considerably differing hash keys, for example, the hash
key for (4,3) is 37.
[0154] From the example described with reference to Table 3, it can
be seen that generating hash keys by "bitwise interleaving" the
binary X and Y coordinates preserves "locality". This example
therefore conceptually illustrates the operation of the bitwise
interleaving spatial hash algorithm that may be used with
representative embodiments of the present invention. However, the
above example is based on the simplified integer based coordinate
system shown in FIG. 8. In order to understand the actual algorithm
that may be used in the implementation of the present control
system, it is necessary to take into account certain other
complexities. These complexities include:
[0155] The fact that GPS and other similar systems which describe
spatial location typically do so using IEEE double-precision
floating-point numbers (not simple integers). For instance, GPS
supplies coordinates in the form of (X,Y) coordinates where X
corresponds to longitude, and Y corresponds to latitude. Both X and
Y are given in units of decimal degrees.
[0156] the fact that certain spatial locations have negative
coordinate values when described using GPS and other similar
coordinate systems. For example, using the WGS84 datum used by
current GPS, the coordinates (153.00341,-27.47988) correspond to a
location in Queensland, Australia (the negative latitude value
indicates southern hemisphere).
[0157] Complexities inherent in representing numbers in accordance
with the IEEE double-precision floating-point numbers standard.
[0158] FIG. 9 shows an example vehicle path similar to that shown
in FIG. 8, except that the coordinates used to describe the points
along the path in FIG. 9 correspond to a "realistic" coordinate
system such as that used by current GPS. In order to understand the
implementation of the bitwise interleaving spatial hash algorithm
when applied to these realistic coordinates, it is necessary to
first appreciate certain aspects regarding the way numbers are
represented using the standard IEEE double-precision floating-point
number format.
[0159] A double-precision floating-point number represented in
accordance with the IEEE 754 standard comprises a string of 64
binary characters (64 bits) as shown in FIG. 10. The number is
represented in three parts, namely the sign, the exponent and the
mantissa. The sign comprises one bit. If the sign bit is 1 then the
number is negative, and conversely if the sign bit is 0 then the
number is positive. The exponent comprises eleven binary
characters, and hence can range from 00000000000 to 11111111111.
However, because of the need to represent numbers that are both
greater and smaller than one, it is necessary to be able to
represent both large positive and large negative values for the
exponent. However, it is not desirable to use one of the exponent
bits to represent the sign of the exponent because this would leave
fewer bits available to represent the exponent's actual value and
would therefore greatly limit the size of the numbers that could be
represented. Therefore, in the IEEE standard 64 bit format, the
true value of the exponent is given by the binary number actually
written by the eleven exponent bits minus an implied exponent
bias.
[0160] Hence, Actual exponent value=written exponent value-exponent
bais
[0161] The exponent bias is 0x3ff=1023. Consequently, the maximum
true exponent value that can be represented (written in decimal
notation) is 1023, and the minimum true exponent value that can be
represented is -1022.
[0162] Finally, the remaining 52 bits form the mantissa. However,
as all non-zero numbers must necessarily have a leading "1" when
written in binary notation, an implicit "1" followed by a binary
point is assumed to exist at the front of the mantissa. In other
words, the leading "1" and the binary point which must necessarily
exist for all non-zero binary numbers is simply omitted from the
actual written mantissa in the IEEE 64-bit standard format. This is
so that an additional bit may be used to represent the number with
greater precision. However, when interpreting numbers which are
represented in accordance with the IEEE standard, it is important
to remember that this leading "1" and the binary point implicitly
exist even though they are not written.
[0163] Bearing in mind these issues, it is possible to understand
the actual spatial hash algorithm used in representative
implementations of the present control system. A "worked" example
illustrating the operation of the spatial hash algorithm to
generate a hash key based on the coordinate (153.0000.degree.,
-27.0000.degree. is given in the form of a flow diagram in FIG. 11.
The points are initially expressed in terms of decimal degrees as
this is the format in which they are delivered from, for example,
GPS.
[0164] From FIG. 11 it can be seen that in order to implement the
algorithm the X and Y coordinates are separated. The next step is
to "normalise" the signs of the respective coordinates (in this
case only the Y coordinate needs to be normalized). The reason for
normalising the signs of the coordinate is because, when
calculating a spatial hash key, it is more convenient to eliminate
negative sign bits from the coordinates. In the case of the
latitude coordinate, those skilled in this area will recognise that
latitude is conventionally written as a number in the range
(-90.degree..ltoreq.latitude.ltoreq.90.degree.). Therefore, by
simply adding 90.degree. to the value of the latitude coordinate,
the spatial hash algorithm can operate with values in the
equivalent "un-signed" or "normalised" latitude range
(0.degree..ltoreq.latitude.ltoreq.180.degree.). Those skilled in
the art will appreciate that the longitude coordinates can also be
normalised to fall within the range (0.degree.
longitude.ltoreq.360.degree.), although that is not necessary in
this example.
[0165] After normalising the coordinates, the next step is to
convert the respective coordinates from their representations in
decimal degrees into binary IEEE double-precision floating-point
number format. This is shown as step 3) in FIG. 11. However, it
will be noted that the binary coordinate representations (and all
other numbers which are generated or used by the algorithm in
binary form) have been written in the alternative hexadecimal
notation for ease of reference and to save space in FIG. 11.
[0166] Next, the binary representations of the two coordinates are
split into their respective exponent (11 bits) and mantissa (52
bits) portions. This is step 4) in FIG. 11. Then, in order to
determine the correct ("true") value of the exponent, the exponent
for each of the coordinate is "de-biased" by subtracting the
implicit exponent bias (0x3ff=1023) as described above. This is
step 5).
[0167] After de-biasing the exponents, the resulting exponents are
then adjusted by a selected offset. The size of the offset is
selected depending on the desired "granularity" of the resulting
fix-point number. In the particular example shown in step 6) of
FIG. 11, the offset is 37, however those skilled in the art will
appreciate this number can be varied to suit.
[0168] After adjusting the exponent, the next step is to
"resurrect" the leading "1" and the binary point which implicitly
exist in the mantissa but which are left off when the mantissa is
actually written (see above). Hence, the leading "1" and the binary
point are simply prepended to the mantissa of each of the
coordinates. This is step 7) in FIG. 11.
[0169] The mantissa for each coordinate is then right-shifted by
the number of bits in the corresponding exponent. The exponents for
each coordinate are then prepended to their corresponding mantissas
forming a single character string for each coordinate. There is
then an optional step of discarding the high-order byte for each of
the two bit fields. This may be done simply to save memory if
required, but is not necessary. Finally, the resultant bit fields
for each coordinate are bitwise interleaved to obtain a single hash
key corresponding to the original coordinates. In the example shown
in FIG. 11, the resultant hash key is 32-bits in length. However,
the length of the resultant hash key may vary depending on, for
example whether the high-order byte is discarded, etc.
[0170] Those skilled in the art will recognise that various other
alterations and modifications may be made to the particular
embodiments, aspects and features of the invention described
without departing from the spirit and scope of the invention.
* * * * *