GB2620940A - Position-based vehicle control scheme - Google Patents

Abstract

Method of autonomously controlling propulsion and braking of a vehicle reaching a halt, by shifting control from a first control scheme using speed sensors to a second predictive control scheme dependent on vehicle forces to minimise stopping position error, if condition is met. The condition may be the vehicle falling below speed threshold of 0.8m/s. Vehicle location may be derived from wheel speed sensors under first control scheme. Speed sensors are not referenced under second control scheme. Vehicle forces may include driving force, braking force, road gradient, or longitudinal acceleration. A comfort constraint (e.g. acceleration or jerk) may be applied. A model of vehicle dynamics may be generated based on vehicle equations of motions to determine a total vehicle force parameter to minimise the stopping position error. An unknown-variable estimator may apply correction to the total force parameter. A position sensor may indicate current vehicle position independent to the speed sensors.

Description

POSITION-BASED VEHICLE CONTROL SCHEME

TECHNICAL FIELD

The present disclosure relates to a control system, a method, and computer software, for implementing a position-based vehicle control scheme. In particular, but not exclusively it relates to transitioning from speed-based control to position-based control when a vehicle is reaching a halt.

BACKGROUND

There are various control schemes for autonomously controlling a vehicle's driving and braking actuators such as an engine and brakes. Typically, the control scheme relies at least in part upon information from one or more speed sensors of the vehicle, such as wheel speed sensors. Wheel speed sensors have quantization errors during low-speed manoeuvring of the vehicle. For example, a wheel speed sensor may read zero speed if vehicle speed is below a threshold such as 0.7km/h and the vehicle moves a short distance. Therefore, a control scheme dependent upon speed measurements may struggle to accurately position the vehicle at very low speeds, resulting in high jerk and/or an inaccurate stop position.

SUMMARY OF THE INVENTION

It is an aim of the present invention to address one or more of the disadvantages associated with the prior art.

Aspects and embodiments of the invention provide a control system, a vehicle, a method, and computer software, as claimed in the appended claims.

According to an aspect of the invention there is provided a control system for autonomously controlling actuators of a vehicle, the actuators comprising a driving actuator and a braking actuator, the control system comprising one or more controllers, the control system configured to: autonomously control the actuators based on a first control scheme dependent on information from one or more speed sensors; determine that a transition condition is satisfied, wherein satisfaction of the transition condition requires at least that the vehicle is reaching a halt; and at least partially switch control of the actuators, based on satisfaction of the transition condition, from the first control scheme to a second control scheme, wherein the second control scheme is a predictive control scheme dependent on information indicative of vehicle forces, wherein the second control scheme is configured to minimise a stopping position error associated with the vehicle reaching said halt.

An advantage is enabling more accurate and comfortable stopping of the vehicle, because of the reduced reliance on reactive information from speed sensors which are subject to quanfisafion errors at low vehicle speeds.

The first control scheme may be configured to minimise stopping position error in dependence on position information determined based on the information from the one or more speed sensors. The one or more speed sensors may comprise wheel speed sensors. By contrast, the second control scheme may be configured to minimise the stopping position error without reference to the one or more speed sensors.

The requirement of the transition condition for the vehicle to be reaching a halt may be implemented using a falling speed threshold. The falling speed threshold may have a value of less than approximately 0.8m/s. This ensures that reactive speed-based control is used in driving scenarios other than low-speed manoeuvring.

The information indicative of vehicle forces may comprise one or more of: measurements indicative of driving actuator force; measurements indicative of braking actuator force; measurements indicative of road gradient; or measurements indicative of longitudinal acceleration. The measurements indicative of longitudinal acceleration may be measurements (calculations) of a derivative of measured speed.

The second control scheme may comprise a comfort constraint based on a derivative of longitudinal velocity. A control action of the second control scheme may be said derivative. An advantage is enabling jerk-limited stopping of the vehicle by limiting predicted jerk as the control action.

The second control scheme may be configured to minimise stopping position error in dependence on a model of vehicle longitudinal dynamics, wherein the model of vehicle longitudinal dynamics is based on vehicle equations of motion and is configured to determine a total vehicle force parameter dependent on a sum of the information indicative of vehicle forces. An advantage is that any constant assumptions underlying such models (e.g., constant road gradient) tend to remain accurate for low-speed, short-distance manoeuvring.

The second control scheme may comprise an unknown-variable estimator to apply a correction to the total force parameter. An advantage is that the performance of the second control scheme improves through online learning.

The second control scheme may be configured to minimise at least stopping position error in dependence on a current vehicle position indicated by information from one or more localisation sensors different from the one or more speed sensors.

According to a further aspect of the invention there is provided a vehicle comprising the control system.

According to a further aspect of the invention there is provided a method of autonomously controlling actuators of a vehicle, the actuators comprising a driving actuator and a braking actuator, the method comprising: autonomously controlling the actuators based on a first control scheme dependent on information from one or more speed sensors; determining that a transition condition is satisfied, wherein satisfaction of the transition condition requires at least that the vehicle is reaching a halt; and at least partially switching control of the actuators, based on satisfaction of the transition condition, from the first control scheme to a second control scheme, wherein the second control scheme is a predictive control scheme dependent on information indicative of vehicle forces, wherein the second control scheme is configured to minimise a stopping position error associated with the vehicle reaching said halt.

According to a further aspect of the invention there is provided computer software that, when executed, is arranged to perform any one or more of the methods described herein. According to a further aspect of the invention there is provided a non-transitory computer readable medium comprising computer readable instructions that, when executed by a processor, cause performance of any one or more of the methods described herein.

BRIEF DESCRIPTION OF THE DRAWINGS

One or more embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings, in which: FIG. 1 illustrates an example of a vehicle; FIG. 2 illustrates an example of a control system; FIG. 3 illustrates an example of a non-transitory computer-readable storage medium; FIG. 4 illustrates an example of a method; FIG. 5 illustrates an example of a Position Predictive Control (PPC) system; FIG. 6 illustrates an example of an Adaptive Position Predictive Control (APPC) system; FIG. 7 illustrates example experimental results comparing PPC and an open-loop control scheme; FIG. 8 illustrates further experimental results comparing PPC and an open-loop control scheme; FIG. 9 illustrates further experimental results comparing PPC and an open-loop control scheme; FIG. 10 illustrates example experimental results comparing APPC and speed-based position control; FIG. 11 illustrates example experimental results comparing PPC and APPC; and FIG. 12 illustrates example experimental results comparing linear Model Predictive Control (MPC) and Optimal Control Problem (OCP) control.

DETAILED DESCRIPTION

FIG. 1 illustrates an example of a vehicle 1 in which embodiments of the invention can be implemented. In some, but not necessarily all examples, the vehicle 1 is a passenger vehicle, also referred to as a passenger car or as an automobile. In other examples, embodiments of the invention can be implemented for other applications, such as commercial vehicles.

FIG. 1 is a front perspective view and illustrates a longitudinal x-axis between the front and rear of the vehicle 1 representing a centreline, an orthogonal lateral y-axis between left and right lateral sides of the vehicle 1, and a vertical z-axis. A forward/fore direction typically faced by a driver's seat is in the negative x-direction; rearward/aft is +x. A rightward direction as seen from the driver's seat is in the positive y-direction; leftward is -y. These are a first lateral direction and a second lateral direction.

FIG. 2 illustrates an example control system 200 configured to implement one or more aspects of the invention. The control system 200 of FIG. 2 comprises a controller 201. In other examples, the control system 200 may comprise a plurality of controllers on-board and/or off-board the vehicle 1.

The controller 201 of FIG. 2 includes at least one processor 204; and at least one memory device 206 electrically coupled to the electronic processor 204 and having instructions (e.g. a computer program 208) stored therein, the at least one memory device 206 and the instructions configured to, with the at least one processor 204, cause any one or more of the methods described herein to be performed. The processor 204 may have an interface 202 such as an electrical input/output I/O for receiving information from speed sensors 214 and other sources 216, 218 and interacting with external components, such as a driving actuator 210 and a braking actuator 212 of the vehicle 1.

A driving actuator 210 of the vehicle 1 comprises, for example, a prime mover such as an internal combustion engine ('engine'), electric machine or the like.

A braking actuator 212 of the vehicle 1 comprises, for example, friction brakes and/or an electric machine configured for regenerative braking. The braking actuator 212 is a different actuator than the driving actuator 210.

The control system 200 is configured to autonomously control the actuators 210, 212. The vehicle 1 may either be a permanently autonomous vehicle, or may have autonomous and non-autonomous (manual) driving modes. In the latter example, the control system 200 may enable user-initiation of autonomous driving functions such as one or more of: an autonomous driving mode; a vehicle parking assistance function (automated pulling into and/or out of parking spaces); an autonomous highway-driving function; a cruise control function; an adaptive cruise control function; an autonomous emergency braking function; and/or the like. Some autonomous driving functions may also enable autonomous control of a steering actuator of the vehicle 1.

FIG. 3 illustrates a non-transitory computer-readable storage medium 300 comprising the instructions (computer software 208).

FIG. 4 is a flowchart illustrating a method 400 according to an example implementation of the invention. The method 400 may be implemented by the control system 200 of FIG. 2. According to an example implementation of the method 400, when it is determined that a vehicle 1 is reaching a halt, a control scheme for autonomously controlling the driving and braking actuators 210, 212 ceases to rely upon information from one or more speed sensors 214 of the vehicle 1, such as wheel speed sensors, and instead relies upon a different control scheme.

This overcomes a low-speed accuracy problem with speed sensors 214. Speed sensors 214 such as wheel speed sensors can have quantization errors. For example, a wheel speed sensor may read zero speed if vehicle speed is below a threshold such as 0.7km/h and the vehicle 1 moves a short distance. Therefore, a control scheme dependent upon speed measurements for position-based control may struggle to accurately position the vehicle 1 at very low speeds, resulting in high jerk and/or an inaccurate stop position.

First, block 402 of the method 400 comprises autonomously controlling the actuators 210, 212 based on a first control scheme dependent on information from one or more speed sensors 214, such as wheel speed sensors. FIG. 2 illustrates one or more speed sensors 214 providing information to the control system 200 In at least some examples, the first control scheme can control both actuators 210, 212 (propulsion and braking) independently.

In an example use case, the first control scheme controls thrust force of the vehicle 1 by controlling the torque of the driving actuator 210 (engine or electric machine), and controls braking force by controlling the pressure at a brake master cylinder of the braking actuator 212 (friction brakes). Other variants where this concept is still applicable include, but are not limited to: -Thrust force delivered by controlling one or a combination of the following: input/output torque of a mechanical/hydraulic transmission, or force/torque delivered at the wheel; -Braking force delivered by controlling one or a combination of the following: braking force/torque at the wheel, braking force/torque generated during regenerative braking (i.e., for hybrid vehicles).

The first control scheme may execute one of the above-described autonomous driving functions. The first control scheme accurately controls the vehicle 1 while the vehicle is moving at an above-threshold speed, because the speed sensors 214 are not subject to quantization errors.

Further, the control system 200 may autonomously stop the vehicle 1, for example, to stop at a traffic light, to stop at a junction, to stop in a queue, to stop in a parking space. In a commercial vehicle use case, the stopping location may be an unloading/loading area, a material handling area, a trailer docking area, a material dumping area, or an excavation area, etc. If the control system 200 determines to, or is instructed to, autonomously stop the vehicle 1, the first control scheme may be used in an initial stopping phase of the stopping manoeuvre while the vehicle 1 is still moving at the above-threshold speed. As described later, a different, second control scheme is used in the final stopping phase of the stopping manoeuvre, as the vehicle 1 reaches a halt.

When stopping the vehicle 1 autonomously, the control system 200 may determine a target stop position at which the vehicle 1 is to stop. The target stop position may comprise, for example, a position within a detected parking space, or a longitudinal separation from a detected obstacle such as another road user. The target stop position may be determined with reference to collision free space information based on information from one or more machine vision sensors (e.g., camera, radar, lidar, etc).

When a target stop position has been determined, the first control scheme is then configured to minimise a stopping position error in dependence on vehicle position information which is determined based on the information from the one or more speed sensors 214.

The stopping position error may be defined as the signed (positive or negative) difference between the vehicle position information (current detected vehicle position) and the target stop position.

The current vehicle position may be determined in dependence on integration of measured wheel speeds. Additional information sources could be used when determining the current vehicle position, such as a vehicle accelerometer, an ultrasonic sensor, or a radar sensor, or a combination thereof, to indicate a relative position to an obstacle.

In some examples, the first control scheme determines a control output (e.g., driving and braking actuator forces) for minimising the stopping position error using a lookup table or reactive feedback controller. The controller determines the control output based on the speed measurements.

In some examples, the first control scheme is dependent on factory calibration data optimised for a range of road gradients and/or vehicle masses. However, the first control scheme may lack robustness to road gradients and/or masses outside the calibrated range.

Block 404 of the method 400 of FIG. 4 determines that a transition condition to switch from the first control scheme to the second control scheme is satisfied, wherein satisfaction of the transition condition requires at least that the vehicle 1 is reaching a halt. The term 'reaching a halt' means that the transition occurs at the final stopping phase of the stopping manoeuvre, in which the vehicle speed is reaching a speed at which speed sensor quantization errors may occur.

The transition condition may be implemented using a falling speed threshold. The falling speed threshold has a value of approximately 0.7km/h (0.2m/s) or another suitable value less than approximately 0.5m/s or less than 0.8m/s. The falling speed threshold may be a wheel speed threshold or a vehicle speed threshold, depending on how speed is measured/calculated.

In another embodiment, the transition occurs upon initiation of the stopping manoeuvre, and the term reaching a halt' refers to the entire stopping manoeuvre.

Block 406 of the method 400 of FIG. 4 comprises autonomously controlling the actuators 210, 212 based on the second control scheme which is now summarised, and several examples of which are later described in detail with comparative experimental results.

The second control scheme is configured to control the actuators 210, 212 to minimise the stopping position error of the vehicle 1 without reference to (or with less reliance on) the one or more speed sensors 214. The second control scheme moves away from a reactive speed-based control scheme to a predictive position-based control scheme.

In order to determine the current vehicle position substantially without reference to the one or more speed sensors 214, the second control scheme may determine the current vehicle position from one or more localisation sensors 216 (FIG. 2) different from the one or more speed sensors 214.

One example localisation sensor 216 is a pixelated imaging sensor (e.g., camera, radar or lidar mounted on the vehicle 1), in the case of landmark-based localisation. Another example localisation sensor 216 is a navigation sensor configured to implement differential positioning (e.g., Differential Global Positioning System, DGPS) or another accurate radionavigation scheme. DGPS has centimetre-scale spatial accuracy, enabling a vehicle 1 to rely upon a DGPS location to perform an accurate manoeuvre.

The second control scheme may control the actuators 210, 212 based on a predictive method utilising information indicative of vehicle forces. For example, the second control scheme may implement a model of vehicle longitudinal dynamics which is based on vehicle equations of motion. The model of vehicle longitudinal dynamics may estimate a total longitudinal vehicle force parameter dependent on a sum of information indicative of individual vehicle forces.

The individual vehicle forces can comprise, for example, thrust forces (e.g., engine forces), braking forces, gravitational forces, and other losses (e.g., aerodynamic drag, road surface changes...). One or more of the vehicle forces, such as braking force, may be treated as a variable, and one or more of the vehicle forces, such as thrust forces, gravitational forces and/or other losses may optionally be treated as a constant in view of the low speed and short distance to be travelled.

It is noted that the second control scheme is not only insensitive to speed sensor quantisation errors, but is also faster and more robust against noise in measurements, since it does not require time-derivation of a signal (as required for speed measurements, resulting in noise amplification).

In order to ensure a comfortable drive, the second control scheme may be configured to satisfy a comfort constraint. The comfort constraint may be based on a second (or first) derivative of longitudinal velocity, such as jerk. The comfort constraint may be measured using a vehicle accelerometer, and predictively controlled by the model of vehicle longitudinal dynamics. In an example implementation, the second control scheme comprises an optimisation function to optimise both the stopping position error and the comfort constraint. Therefore, an accurate and jerk-limited stopping manoeuvre is achieved.

The predictive nature of the second control scheme enables the calculation of optimal control actions even with very little data history. On the contrary, speed-based control schemes (the first control scheme) are based on reactive algorithms (e.g., PID or sliding mode control) that require several seconds to build up integration errors and filter input signals to correctly estimate the derivative errors. For this reason, the second control scheme is suitable to be used on vehicles with slow actuator dynamics (such as internal combustion engines) that have to be controlled in a short time span. In situations such as stopping the car to a zero speed, a manoeuvre that can last less than 1 second, there is often not enough time to build up integration errors or filter correctly the derivative error. For these reasons reactive controllers are not used in the second control scheme.

The second control scheme may only be required for a short amount of time. The vehicle 1 may be autonomously driven for the majority of a desired path via speed-based control (e.g., the first control scheme), and the vehicle 1 has to stop precisely at stopping points. The control system 200 may switch from speed control to position control when the car is reaching zero speed. The predictive nature of the second control scheme enables position control of the vehicle 1 to be fast in reacting to the current vehicle position and in ensuring an adequate vehicle final stop.

Additionally, in some examples of the disclosure the second control scheme is configured to adaptively learn and compensate for the effect of unmeasured external disturbances, such as: road gradient changes; vehicle mass changes; road type changes (gravel, grass, etc); and tyre characteristics (adherence and size); and/or vehicle pose (suspension state).

In the following description, specific examples of the second control scheme are described. Non-adaptive and adaptive versions are described. The non-adaptive version is referred to as Predictive Position Control (PPC). The adaptive version is referred to as Adaptive Predictive Position Control (APPC). Both PPC and APPC are targeted at controlling the longitudinal actuators (driving actuator 210 and braking actuator 212) of the vehicle Ito stop the vehicle 1 at a target stop position.

The second control scheme utilises a model-based predictive optimal control. Different control algorithms can be used to calculate the optimal actuator control actions needed to stop the vehicle 1 at the target stop position, while minimizing the jerk (i.e., time derivative of the acceleration) or acceleration of the manoeuvre. Two example control algorithms are described herein: a Linear Model Predictive Control (MPC); and a Non-Linear Model Predictive Control also known as Optimal Control Problem (OCP).

FIG. 5 sets out an example system diagram of a PPC system 500. The system comprises a PPC controller 502. The PPC controller 502 is an example implementation of the second control scheme, and includes a model of vehicle longitudinal dynamics. The PPC controller 502 receives inputs for the model from various sources. The PPC controller's outputs are transmitted to the driving actuator 210 and the braking actuator 212.

The illustrated inputs to the PPC controller 502 of FIG. 5 comprise: - a reference trajectory information source 504, such as a trajectory planner function of a controller, to indicate a jerk-limited deceleration profile from a crawl speed (e.g., 1km/h) to stationary (e.g., Okm/h), satisfying a jerk constraint (e.g., 1m/s^3 limit) ; -a measured vehicle position information source 506, which may be a localisation sensor 216 (FIG. 2); - an initial inertial forces information source 508, such as the speed sensor and/or an accelerometer, to indicate initial conditions of the model; -a road gradient information source 510, such as a vehicle inertial measurement unit (IMU) 218 (FIG. 2) (e.g., utilising one or more accelerometers/gyroscopes to indicate at least vehicle pitch); and -a vehicle dynamics base data information source 512, providing a-priori information relating to vehicle dynamics of the vehicle 1, such as information that cannot be sensed onboard the vehicle 1.

The PPC controller 502 determines the longitudinal distance to the target stop position (the stopping position error) based on measurements from the reference trajectory information source 504 and from the measured vehicle position information source 506. Block 518 represents vehicle movement, and is therefore connected to source 506 as a feedback loop.

The PPC controller 502 determines initial conditions for the model such as longitudinal vehicle speed and longitudinal acceleration, based on measurements from the initial inertial forces information source(s) 508.

Being a model-based approach, the PPC may receive from source 512 a-priori vehicle dynamics base data information relating to vehicle dynamics of the vehicle 1, such as vehicle mass data (e.g., kerb weight or gross vehicle weight), powertrain loss data, and/or aerodynamic drag data (e.g., drag coefficient, frontal area). These parameters may be stored from factory in a vehicle memory or communicated wirelessly to the vehicle via Vehicle-to-X (V2X) infrastructures. It would be appreciated that in some implementations, at least some of this information could instead be measured via onboard sensors.

A first solution to the PPC control problem is achieved with a linear Model Predictive Control (MPC). It would however be understood that MPC is one of many predictive control schemes that could implement the invention (others include LQRs, Constrained MPC, and Non-Linear MPC). However, MPC has several advantages as described below.

A significant advantage of the MPC over other approaches is that it can calculate the optimal control action algebraically, by calculating a set of matrices multiplications. This means that, for a given model and set of initial and final conditions, the calculated optimal solution will always be the best one (the optimization will reach a global minima). The implication of this is important since this guarantees that the behaviour of the MPC will be always deterministic and predictable (neglecting small variations of the solution that might be due to numerical rounding errors while doing matrices operations). Having a deterministic and predictable control scheme means that this software can be applied to edge cases that may be encountered during driving.

On the other hand, MPC requires the system to be controlled to be linear (which is the case in this disclosure) and MPC may find it difficult to handle hard constraints (whereas the later-described OCP can handle hard constraints).

Before proceeding, it is worth mentioning the different types of constraints used in the predictive control schemes described herein: 1) Hard constraints: these can be applied to the control action or the control states, and represent limits that should never be exceeded. As an example, the vehicle speed can be a control state that should never become negative to avoid roll-back.

2) Soft constraints: these are constraints that can be violated but if this happens, the current calculated solution will be penalized over others, hence will not be the optimal one. These constraints form part of the target function (e.g., cost function) to be minimized by the predictive controller. As instance, the MPC implementation described below targets zero longitudinal jerk, so a zero-jerk requirement is implemented as a soft constraint (comfort constraint). By doing so, the target function will have two terms: a) to minimize the stopping position error (primary target), while minimising the longitudinal jerk (soft constraint/secondary target). Hence, the target function will discard automatically all the solutions that have a large longitudinal jerk and prefer the ones with a low longitudinal jerk.

As mentioned, MPC cannot handle hard constraints, but it can handle soft constraints. This implies that the MPC will be able to handle smooth stopping manoeuvres with low initial deceleration, while for aggressive stops with large initial decelerations this may be more difficult (but solvable by the later-described OCP approach). It has to be noted however that smooth stopping manoeuvres represent the majority of the scenarios experienced in an autonomous vehicle, and for this reason the MPC is an advantageous control scheme over others (like the OCP described in the next section).

The MPC -both linear and non-linear -comprises a model of the vehicle longitudinal dynamics. In this example a longitudinal bicycle model is used, and it is formulated as: fa = t = v F ot = a. = it

where s is the longitudinal position of the vehicle 1, v is the longitudinal speed, a is the longitudinal acceleration, d is the longitudinal jerk (time derivative of acceleration a), m is the vehicle mass, and u is the control action.

Plot is the sum of all the forces acting on the vehicle 1 longitudinally: Ftot = Feng Fgrav -Floss -Fbrake wherein: -Feng is a measurement indicative of driving actuator force (the measured force of the driving actuator 210, e.g., force of the engine projected at the wheel); -Fgrav = m*g*sin(RoadGradient) is the measured gravitational force on the vehicle 1 (positive when going downhill forward or uphill in reverse) based on a measurement indicative of road gradient (e.g., from source 510, FIG. 5); -Floss is an unknown-variable estimator to apply a correction to the total force parameter based on unmeasured losses (i.e., estimated losses on the powertrain and the aerodynamic drag, e.g., from source 512, FIG. 5); and &brake is a measurement indicative of braking actuator force (the measured force of the braking actuator 212, e.g., brake torque projected at the wheel).

To simplify this formulation, PPC controller 502 can assume that Feng, Fgrav and Floss are constant.

This can be justified by the fact that the final stopping phase of the vehicle 1 is a manoeuvre that typically last few seconds (e.g., no more than 2s) and the longitudinal vehicle dynamics are usually slow if compared to the dynamics of the measurements acquired (sensor noise).

In the formulation above, the control action u is the ratio between the time derivative of Hot and the vehicle mass. In other words, the control action is the time derivative of the acceleration ci = u and this allows the MPC to control the longitudinal movements of the car in terms of longitudinal jerk.

In State Space (SS) form the above dynamics become: it = Ax Bu y = Cx + Du where = [s and the SS matrices in a time-discrete formulation are: 1 T 0 A =Fi) 1 T 8= 01 [0 0 0 1 C = [1 0 0] D= 0] Note that by having C = [1 0 0], it means that the only state that has to be considered in the target function for the MPC is the longitudinal position of the vehicle 1, and since the other states (speed and acceleration) are not needed, the formulation of the MPC is significantly simplified and can be calculated requiring less computational resources. The solution of the predictive control scheme then becomes efficient and can be run on hardware with limited computing power.

The final component to complete the formulation of the MPC is the target function: = (. - )2 (2mint""J w1.S N2 ut4 The MPC will minimize the function J which is a weighted sum of two quadratic terms. the distance to stop, and the energy of the control action (i.e. the longitudinal jerk) that is used as mentioned earlier as a soft constraint. The MPC therefore minimises the longitudinal jerk of the car (i.e. optimize for comfort) by minimising the energy of the control action.

Once the control action has been calculated, it is possible for the MPC to obtain the positive and negative actuator forces (Feng and Fbrake respectively in this example) as follows: -Positive control action: Feng is calculated assuming that the driving actuator 210 has to be able to at least compensate the effect of Fgrav and Floss: Feng = (-Fgrav + Floss) * It k is a small percentage increment (>=1) that might be added to take into account measurements and estimation uncertainty.

-Negative control action: the predicted longitudinal position s,,,p, is numerically differentiated twice in time to obtain the predicted longitudinal acceleration ampG, then: Fbrake = Feng Fgrav -Floss -inamp, A second solution to the PPC control problem is the Optimal Control Problem (0CP, also known as non-linear MPC). OCP is proposed as a more flexible yet computationally demanding control strategy. As shown in later results, this OCP control strategy is more suitable to be used on aggressive stops (with large initial decelerations) because it can handle soft and hard constraints. This property of the OCP ensures that, when an optimal solution is calculated, it is feasible and achievable by the vehicle 1 (i.e., it will not request negative brake pressures).

Unlike the MPC, that is always able to find an optimal result, but this solution might be unfeasible for the vehicle 1, the OCP does not guarantee to calculate an optimal solution but when it does make a calculation, the solution will be feasible by the vehicle 1. Whether the OCP might be able to calculate an optimal solution depends on: 1) Hard constraints: when these are violated the OCP will not converge to an optimal solution. This implies that care has to be put in defining constraints and ranges on states and control actions, to minimise the chance of having an OCP not converging to a solution.

2) Initial conditions: the OCP calculates the optimal control action not algebraically (like the MPC does) but numerically, with an optimisation algorithm that requires multiple iterations to find the minimum of the target function. Depending on how far the initial condition is from the desired minimum, and depending on the chosen optimisation scheme, the OCP might take too many optimisation iterations and might not converge to the minimum in an acceptable time (maximum time may be set by the sampling rate of the control software).

The OCP is able to calculate more complex control profiles that are suitable to handle challenging braking manoeuvres (i.e., with large initial decelerations).

The OCP has some elements in common with linear MPC: the target function; dynamics of the system to be controlled; and boundary conditions. Additionally, the OCP can handle non-linear target functions, non-linear system dynamics, and hard constraints.

The formulation of the OCP is the following: -Target function, can be identical to the one used for the MPC: = v v 1(5(0 -sun.

get)2 "72 -Dynamic model, can be identical to the one used for the MPC m K = Fen, + Fyn, -Fins, -Fryrk = k -FbrkF = -tot -Initial and final conditions, the former are identical to the ones used for MPC, while the latter are new and treated as hard constraints by the OCP: s(to) = 0 s(t1) = Free v(to) = I/o (Fixed) v(t) = 0 Feee(to) = (Fixed) Fene(tr) = 0 u(to) = Uo (Fixed) u(t1) = Free - Hard constraints: 0 < s(t) smax 0 < a(t) < Vo with Vo > 0 Ftut,"th, (FbrakMax) F10(t) Ftr,t,,,,m"(FbrakeM in) uMin < u(t) < uMax This OCP formulation represents a non-linear optimization problem with differential-algebraic constraints (the dynamics and limits for states/control action) whose solution is not trivial. Solving the OCP can rely upon: direct methods (the one used to calculate the solution in the results described later); indirect methods; multiple-shooting methods; dynamic programming; or the spectral method.

As an improvement to PPC, the second variant, the APPC (FIG. 6), comprises a learning engine configured to automatically learn the vehicle dynamics (i.e. mass, losses, and other disturbances due as instance to sensor miss-calibration) while the vehicle 1 is being driven autonomously. This significantly eases the calibration process of the position control, since the control parameters will be automatically learnt.

FIG. 6 illustrates an example system diagram of an APPC system 600. The difference compared to FIG. 5 is that the vehicle dynamics base data information source 512 is omitted, and instead a learning engine 602 is provided which comprises a sensor fusion block 6020 and an inertia and loss calculation block 6022.

The sensor fusion block 6020 is configured to fuse data from inertial sensors (e.g., accelerometers/IMU 218) with the vehicle ego-state (e.g., driving actuator output such as engine torque, braking actuator output such as brake pressure, and vehicle speed whether sensed by speed sensors or by other means) and the measured road gradient (e.g., pitch angle estimated from the Inertial Measurement Unit 218).

The model of vehicle longitudinal dynamics may be a dynamic model, calculated in real-time using the sensor fusion. A Kalman filter or equivalent can be used to estimate the losses in real time.

Some of the advantages in estimating the vehicle dynamics in real time are: -automatically adapting to non-linear behaviours that cannot be described by the linear dynamics model of the MPC and OCP formulation (e.g. linear Floss term); - automatically adapting to uncertainty in the model parameters (e.g. vehicle mass changes); -automatically calculating the losses (Floss), without requiring dedicated identification tests for each vehicle; and - automatically adapting to measurement uncertainty (e.g. offsets on measured road gradient, or noise in the measured engine force Fang).

The inputs to the Kalman Filter include Fbrake, Feng, Fgrav, the measured travelled distance of the vehicle 1, the measured speed of the vehicle 1, and the measured acceleration of the vehicle 1 (the latter two calculated based on the distance travelled). An example formulation of the Kalman Filter is as follows: The control action is the derivative of the loss force, and it is calculated as follows: FEng -Fbrake + FGrav in In State Space (SS) form the above dynamics become: = Ax Bu ty = Cx + Du where y = [s and the SS matrices in a time-discrete formulation are identical to the ones used for the MPG in the 1 T 0 01 A = [0 1 T1 B = FO 0 0 1 C = [1 0 0 D = [0] The goal of the Kalman filter is to calculate the longitudinal acceleration a"t KF, which is passed to the inertia and loss calculation block 6022. The inertia and loss calculation block 6022 calculates the force of inertia based on the calculated longitudinal acceleration, and then an estimated loss force Floss based on the force of inertia: Finertia = in a --est_KF Floss estimated = -Fbrake + Feng -Fgrav -Finertia The following FIGS. 7-12 refer to experimental vehicle-stopping results graphs. The results include: -in FIGS. 7 to 9, comparison between PPC and an open-loop solution with constant brake pressure; -in FIG. 10, comparison between APPC and speed-based position control (such as the first control scheme); -in FIG. 11, comparison between PPC and APPC; and UK? = PPC: -in FIG. 12, comparison between linear MPC and OCP.

FIG. 7 shows an accuracy comparison when a stop manoeuvre is done by a constant brake pressure (left), compared to when the manoeuvre is done by PPC (MPC method) in a low gradient surface (right).

In the top graphs, y=Distance to Stop signal in meters and x=fime. In the bottom graphs, y= brake pressure in bar and x=time (aligned with top graphs).

There are four stops that are identified when the Distance to Stop signal in the top graphs changes towards a zero value (dotted line). When the stopping point is an overshoot (signal below the dotted line), the vehicle 1 has stopped after the stopping point. On the other hand, when the vehicle 1 is not reaching zero (signal above dotted line), it has stopped before the stopping point.

In the experiments done with stops in open loop, the brake pressure was fixed to 8 bar because this is a value has been found to be comfortable with in-vehicle calibration.

FIG. 7 shows that the level of accuracy and precision in the open loop solution is lower and outside the specification for most use cases such as parking, even when the solution in open loop has been calibrated for the specific conditions.

FIG. 8 presents a box plot comparison of the final stopping position errors in metres (y-axis). The open loop solution is in the left box plots and the PPC solution of FIG. 7 is in the right box plots.

Within each set of box plots, the boxes are (from left to right): downhill:forward; downhill:reverse; flat:forward; and uphill:forward, referring to the road gradient and direction of travel of the vehicle 1.

The study was prepared following the same route (one route in a flat road and one route in a higher gradient road) several times for both cases.

The illustrated improvement results for PPC are: -71.0% in slopes -Average of 21cm with PPC -56.3% in flat -Average of 13 cm with PPC FIG. 9 shows an accuracy and smoothness comparison when a stop manoeuvre is done by a constant brake pressure (left), compared to when the manoeuvre is done by the PPC (right) scheme used in FIGS. 7-8.

In the top graphs, y=stopping position error in metres and x=maximum longitudinal acceleration. In the bottom graphs, y= stopping position error in metres and x=maximum longitudinal jerk.

The PPC approach is seen to consistently provide a better optimisation of acceleration/jerk and stopping accuracy.

In FIG. 10, stopping via APPC is compared with stopping via the reactive speed-based position control scheme ('first control scheme' of FIG. 4). The speed-based position control solution is in the left box plots and the APPC solution is in the right box plots. Otherwise, the box plots are presented and ordered in the same manner as FIG. 8.

When compared to the first control scheme, it is possible to appreciate how the adaptation and prediction terms of the APPC provide a significant improvement in controlling the vehicle 1 on challenging situations: on average on flat surfaces APPC is more repeatable and precise (by 146%, from average 32cm to just average 13cm) and on slopes the improvement is even larger (approx. 500%, from average 126cm to just average 21cm).

In FIG. 11, PPC (left) and APPC (right) are compared with each other. The test was performed including high gradient roads of almost 10 degrees. The graphs are presented and ordered in the same manner as FIG. 7 (distance to stop on top, brake pressure on bottom).

The biggest improvement of APPC is detected in reverse, which is the last stop (indicated in a thick square in FIG. 11). APPC does not require the calibration of separate forward and reverse models of losses. APPC learns automatically and in real-time those losses.

From this comparison, it is observed that having an adaptive part on the Predictive Position Control provides the benefit that it simplifies the calibration of the software, since in the pure Predictive Position Control the vehicle dynamics have to be identified for several conditions, while for the Adaptive Predictive Position Control the software will auto-adapt to changing conditions.

FIG. 12 illustrates graphs comparing the performance of linear MPC (left) and OCP (non-linear MPC) (right) on a flat surface with a low initial acceleration of -0.5m/s2. All of the graphs have aligned time axes (x-axis). From top to bottom, the y-axes represent: distance (m) (dashed line = target); vehicle speed; vehicle acceleration; brake pressure; and control action (brake pressure per second, balls).

The other dashed lines in the linear MPC graphs represent ultimate targets (e.g., travelled distance of 1.5m, zero speed and zero acceleration).

The other dashed lines in the OCP graphs represent hard constraints.

FIG. 12 shows that both linear MPC and OCP can stop the vehicle 1 within the prescribed target, with a low value of the control action (limited to approximately 20 bads).

In further testing for higher initial acceleration (-1.5m/s2), OCP outperformed linear MPC since it could change the brake pressure faster. However, additional heuristics can improve the performance of linear MPC without needing to implement OCP. This is because heuristics in conjunction with a continuous solution of the linear MPC will define at every calculation step a set of initial conditions that are different to the ones predicted by the MPC but always feasible. Since at every software execution cycle the initial conditions are the measured values (not theoretical), the first points in the calculated control horizon will be always feasible. The manoeuvre will then be adjusted before reaching the unfeasible solutions. This last point allows the MPC to be used in reality on aggressive manoeuvres, accepting that in those cases the calculated trajectory might be sub optimal and less comfortable.

For purposes of this disclosure, it is to be understood that the controller(s) described herein can each comprise a control unit or computational device having one or more electronic processors. A vehicle and/or a system thereof may comprise a single control unit or electronic controller or alternatively different functions of the controller(s) may be embodied in, or hosted in, different control units or controllers. A set of instructions could be provided which, when executed, cause said controller(s) or control unit(s) to implement the control techniques described herein (including the described method(s)). The set of instructions may be embedded in one or more electronic processors, or alternatively, the set of instructions could be provided as software to be executed by one or more electronic processor(s). For example, a first controller may be implemented in software run on one or more electronic processors, and one or more other controllers may also be implemented in software run on one or more electronic processors, optionally the same one or more processors as the first controller. It will be appreciated, however, that other arrangements are also useful, and therefore, the present disclosure is not intended to be limited to any particular arrangement. In any event, the set of instructions described above may be embedded in a computer-readable storage medium (e.g., a non-transitory computer-readable storage medium) that may comprise any mechanism for storing information in a form readable by a machine or electronic processors/computational device, including, without limitation: a magnetic storage medium (e.g., floppy diskette); optical storage medium (e.g., CD-ROM); magneto optical storage medium; read only memory (ROM); random access memory (RAM); erasable programmable memory (e.g., EPROM and EEPROM); flash memory; or electrical or other types of medium for storing such information/instructions.

It will be appreciated that various changes and modifications can be made to the present invention without departing from the scope of the present application.

The blocks illustrated in FIGS. 4-6 may represent steps in a method and/or sections of code in the computer program 208. The illustration of a particular order to the blocks does not necessarily imply that there is a required or preferred order for the blocks and the order and arrangement of the block may be varied. Furthermore, it may be possible for some steps to be omitted.

Although embodiments of the present invention have been described in the preceding paragraphs with reference to various examples, it should be appreciated that modifications to the examples given can be made without departing from the scope of the invention as claimed.

Features described in the preceding description may be used in combinations other than the combinations explicitly described. Although functions have been described with reference to certain features, those functions may be performable by other features whether described or not. Although features have been described with reference to certain embodiments, those features may also be present in other embodiments whether described or not.

Whilst endeavouring in the foregoing specification to draw attention to those features of the invention believed to be of particular importance it should be understood that the Applicant claims protection in respect of any patentable feature or combination of features hereinbefore referred to and/or shown in the drawings whether or not particular emphasis has been placed thereon.

Claims (15)

1. CLAIMS1. A control system for autonomously controlling actuators of a vehicle, the actuators comprising a driving actuator and a braking actuator, the control system comprising one or more controllers, the control system configured to: autonomously control the actuators based on a first control scheme dependent on information from one or more speed sensors; determine that a transition condition is satisfied, wherein satisfaction of the transition condition requires at least that the vehicle is reaching a halt; and at least partially switch control of the actuators, based on satisfaction of the transition condition, from the first control scheme to a second control scheme, wherein the second control scheme is a predictive control scheme dependent on information indicative of vehicle forces, wherein the second control scheme is configured to minimise a stopping position error associated with the vehicle reaching said halt.
2. 2. The control system of claim 1, wherein the first control scheme is configured to minimise stopping position error in dependence on position information determined based on the information from the one or more speed sensors.
3. 3. The control system of claim 2, wherein the one or more speed sensors comprise wheel speed sensors.
4. 4. The control system of claim 2 or 3, wherein the second control scheme is configured to minimise the stopping position error without reference to the one or more speed sensors.
5. 5. The control system of any preceding claim, wherein the requirement of the transition condition for the vehicle to be reaching a halt is implemented using a falling speed threshold.
6. 6. The control system of claim 5, wherein the falling speed threshold has a value of less than approximately 0.8m/s.
7. 7. The control system of any preceding claim, wherein the information indicative of vehicle forces comprises one or more of: measurements indicative of driving actuator force; measurements indicative of braking actuator force; measurements indicative of road gradient; or measurements indicative of longitudinal acceleration.
8. 8. The control system of any preceding claim, wherein the second control scheme comprises a comfort constraint based on a derivative of longitudinal velocity.
9. 9. The control system of claim 8, wherein a control action of the second control scheme is said derivative.
10. 10. The control system of any preceding claim, wherein the second control scheme is configured to minimise stopping position error in dependence on a model of vehicle longitudinal dynamics, wherein the model of vehicle longitudinal dynamics is based on vehicle equations of motion and is configured to determine a total vehicle force parameter dependent on a sum of the information indicative of vehicle forces.
11. 11. The control system of claim 10, wherein the second control scheme comprises an unknown-variable estimator to apply a correction to the total force parameter.
12. 12. The control system of any preceding claim, wherein the second control scheme is configured to minimise at least stopping position error further in dependence on a current vehicle position indicated by information from one or more localisation sensors different from the one or more speed sensors.
13. 13. A vehicle comprising the control system of any preceding claim.
14. 14. A method of autonomously controlling actuators of a vehicle, the actuators comprising a driving actuator and a braking actuator, the method comprising: Autonomously controlling the actuators based on a first control scheme dependent on information from one or more speed sensors: determining that a transition condition is satisfied, wherein satisfaction of the transition condition requires at least that the vehicle is reaching a halt; and at least partially switching control of the actuators, based on satisfaction of the transition condition, from the first control scheme to a second control scheme, wherein the second control scheme is a predictive control scheme dependent on information indicative of vehicle forces, wherein the second control scheme is configured to minimise a stopping position error associated with the vehicle reaching said halt.
15. 15. Computer software that, when executed, is arranged to perform a method according to claim 14.