SYSTEM FOR CONTROLLING LATERNAL MOVEMENT OF AN AUTONOMOUS MOTOR VEHICLE

Abstract

A system for controlling the lateral movement of an autonomous vehicle (101) is disclosed, comprising: a trajectory planner (102) configured to generate a target trajectory for the vehicle (101); and a trajectory controller (104) configured to receive the target trajectory from the trajectory planner (102) and output a steering control signal for controlling the lateral movement of the vehicle (101). The trajectory controller (104) includes a control circuit (104a), wherein the control circuit (104) comprises an integrated adaptive PID neural network controller (124).The integrated adaptive PID neural network controller (124) comprises: a PID controller (116) that receives an input signal dependent on a target lane position parameter of the target trajectory and is configured to minimize an error value that specifies the difference between the actual trajectory of the vehicle (101) and the input signal dependent on the target lane position parameter, wherein the PID controller (116) is further configured to receive an input signal dependent on a target heading angle parameter of the target trajectory and is configured to minimize an error value that specifies the difference between the actual trajectory of the vehicle (101) and the input signal dependent on the target heading angle parameter; and a neural network (118) that is configured to determine the P-gain, I-gain, and D-gain terms used by the PID controller (116).The integrated adaptive PID neural network controller (124) is configured to use the P and I gain terms provided by the neural network (118) to control the lane position of the vehicle (101) and the D gain term provided by the neural network (118) to control the heading angle of the vehicle (101), whereby the P, I and D gain terms used by the PID controller (116) are provided by the same neural network (118).

Description

[0001] The present application relates to a system for controlling lateral movement of an autonomous motor vehicle and to an autonomous vehicle comprising the system for controlling lateral movement.

[0002] To control an autonomous vehicle so that it can drive with reduced or no user input, a target trajectory is determined. The target trajectory can be determined by a planner function that processes images of the vehicle's location and surroundings along with other sensor information available to the vehicle to calculate a target trajectory. The target trajectory is the desired trajectory of the vehicle so that it follows the road or performs a desired maneuver, such as changing lanes, turning onto another road, or taking an evasive maneuver to avoid a collision.

[0003] Once a target trajectory is determined for the vehicle, the vehicle's steering, along with other vehicle devices, must be controlled to maintain the trajectory. Therefore, a lateral movement control system is provided that can implement a feedback loop in which the vehicle's current trajectory is compared with the target trajectory. A control signal is provided to the vehicle's steering system to minimize the difference between them during the vehicle's lateral movement maneuvers.

[0004] Known motion control systems can use feedback loops based on fixed parameters, which can become suboptimal for vehicle control over time. This can result in inaccurate tracking of the target trajectory or a longer time to reach the target trajectory. Adjusting the vehicle's steering angle to achieve the desired lateral trajectory, or the actual steering path, is challenging, as it is subject to several variable or unknown influences, such as road contact, vehicle load, vehicle tires, etc.

[0005] The present application relates to a lateral movement control system for autonomous motor vehicles, which aims to improve the limitations of the known control systems described above.

[0006] In a first aspect, the present application provides a system for controlling the lateral movement of an autonomous vehicle, comprising: a trajectory planner configured to generate a target trajectory for the vehicle; and a trajectory controller configured to receive the target trajectory from the trajectory planner and generate as output a steering control signal for controlling the lateral movement of the vehicle, wherein the trajectory controller comprises a control circuit, wherein the control circuit comprises one or more adaptive neural network controllers, each comprising: a feedback loop controller (e.g., PID type) receiving an input signal dependent on a parameter of the target trajectory and configured to minimize an error value indicative of the difference between the actual trajectory of the vehicle and the input signal based on a P-gain and / or I-gain and / or D-gain term; and a neural network configured to determine a P-gain and / or I-gain and / or D-gain term used by the feedback loop controller.

[0007] By providing gain terms calculated by a neural network, the operation of the feedback loop control can be adjusted or tuned online and in real time while the motion control system is running. This is advantageous compared to many existing control systems that use fixed gain terms and can improve the accuracy and effectiveness of vehicle control.

[0008] The feedback loop controller can use all three gain terms, P, I, and D, and thus be a PID controller. Alternatively, the feedback loop controller can use only one or two of the gain terms, and thus be a P, I, D, PD, or PI controller.

[0009] The target trajectory may include one or more target trajectory parameters and / or variables. The target trajectory parameters and / or variables may include one or more target curves, a target lane position, and a target heading angle.

[0010] The trajectory controller may receive one or more of the target trajectory parameters or signals dependent thereon as input. The trajectory controller may also receive one or more signals related to the vehicle's current trajectory.

[0011] The one or more adaptive neural network controllers for lateral trajectory control may include an adaptive neural network controller for lane position configured to perform P and / or I and / or D control for the lane position of the trajectory of the vehicle.

[0012] The one or more adaptive neural network controllers may include an adaptive heading angle neural network controller configured to perform P and / or I and / or D control of the heading angle of the trajectory of the vehicle.

[0013] The adaptive neural network controller for lane position may be configured to receive a target lane position parameter as input and generate an output control signal. The adaptive neural network controller for heading angle may be configured to receive a sum of the output control signal of the adaptive neural network controller for lane position and a target heading angle parameter as input. This allows the lane position and heading angle controllers to be connected in series.

[0014] The trajectory controller may further comprise a turnaround PID controller configured to provide P and / or I and / or D control of the turnaround. The turnaround controller may use P, I and / or D gain terms that are predetermined and / or dependent on the vehicle motion variable(s). The turnaround PID controller may use P, I and D gain terms that are configured to be dependent on the vehicle motion variable(s), such as vehicle speed and heading turnaround measurements. The turnaround controller may use gain terms derived from a lookup table of predetermined values ​​or use a fixed functional relationship between gain values ​​and vehicle motion variables.The gain values ​​are therefore predetermined in the sense that there is a fixed relationship between a current vehicle motion variable and the gain term(s) P, I, and D. Therefore, the curve control may not use a neural network to provide the gain terms.

[0015] Alternatively, the waveform controller may be an adaptive neural network controller configured to provide the required P-gain, I-gain, and D-gain terms.

[0016] The curve control may be configured to receive a target curve parameter as input and generate a curve control signal as output. The trajectory control may be configured to determine the sum of the curve control signal and the output of the adaptive neural network heading angle control (when connected in series with the lane position control) to provide the trajectory control's steering control signal output. This allows the curve control to be arranged in parallel with the other two controllers. This allows the curve control to be incorporated earlier in the steering calculation and improves the accuracy of the steering control signal.

[0017] Each of the neural networks in the trajectory controller may include an input layer of neurons, at least one hidden layer of neurons, and an output layer with at least one output neuron. Each neural network may receive input neurons, the error value associated with a respective trajectory parameter, and the control signal output by the respective controller, and output the gain values ​​at respective node values ​​within the hidden layer of the neural network.

[0018] Each of the neural networks may further be configured to receive at least one environmental variable as a feedforward term.

[0019] The environmental variable may include at least one of the following variables commonly found in a vehicle motion control system: - the speed of the vehicle for most of the vehicle movement control, - Road / traffic measurements such as curves, direction of travel, speed limits, etc. - Vehicle acceleration / reaction force - Motor speed / angle for current / torque control - Steering torque or angle - Vehicle indicator status, e.g. turning indicated or turning not indicated

[0020] During use of the control system, the weights of the neural network(s) can be updated by a gradient descent backpropagation method each time a new set of input values ​​is fed to the neural network, and the updated weights are used together with the input values ​​to update the neurons.

[0021] In a second aspect, the present application provides a vehicle comprising the motion control system of the first aspect.

[0022] According to a third aspect, there is provided a system for controlling lateral movement of an autonomous vehicle, comprising: a trajectory planner configured to generate a target trajectory for the vehicle; and a trajectory controller configured to receive the target trajectory from the trajectory planner and generate as an output a steering control signal for controlling the lateral movement of the vehicle; wherein the trajectory controller comprises a control circuit, wherein the control circuit comprises an integrated adaptive PID neural network controller comprising: a PID controller that receives an input signal dependent on a target lane position parameter of the target trajectory and is configured to minimize an error value indicative of the difference between the actual trajectory of the vehicle and the input signal dependent on the target lane position parameter, wherein the PID controller is further configured to receive an input signal dependent on a target heading angle parameter of the target trajectory and is configured to minimize an error value indicative of the difference between the actual trajectory of the vehicle and the input signal dependent on the target heading angle parameter; and a neural network configured to determine the P-gain, I-gain and D-gain terms used by the PID controller, wherein the integrated adaptive PID neural network controller is configured to use the P and I gain terms provided by the neural network to control the lane position of the vehicle and the D gain term provided by the neural network to control the heading angle of the vehicle, whereby the P, I, and D gain terms used by the PID controller are provided by the same neural network.

[0023] Using a single PID controller and a single neural network to control both lane position and heading angle can improve the efficiency of the control system. This also reduces the number of control steps and the number of required components compared to using separate dedicated PID controllers and / or neural networks.

[0024] The trajectory controller may further comprise a curve control configured to perform P and / or I and / or D control of the curve of the trajectory of the vehicle.

[0025] The turn-path controller may use P, I, and / or D gain terms that are predetermined and / or dependent on the vehicle motion variable(s). The turn-path controller may therefore not use a neural network to provide its gain terms. The turn-path PID controller may use P, I, and D gain terms that are designed to be dependent on the vehicle motion variable(s), such as vehicle speed and heading turn-path measurements. The turn-path controller may use gain terms derived from a lookup table of predetermined values ​​or use a fixed functional relationship between gain values ​​and vehicle motion variables. The gain values ​​are therefore predetermined in the sense that there is a fixed relationship between a current vehicle motion variable and the P, I, and D gain term(s).

[0026] Alternatively, the waveform controller may be an adaptive PID neural network controller configured to provide its required P-gain, I-gain, and D-gain terms.

[0027] The curve control may be configured to receive a target curve parameter as input and generate a curve control signal as output. The trajectory control may be configured to determine the sum of the curve control signal and the output of the integrated adaptive PID neural network controller to provide the steering control signal output of the trajectory control. This allows the curve control and the integrated controller to be connected in parallel. This allows the curve control to be incorporated earlier in the control calculation.

[0028] The neural network(s) may comprise an input layer of neurons, at least one hidden layer of neurons, and an output layer with at least one output neuron. The neural network(s) may comprise a feedforward neural network that outputs the error value associated with the input trajectory parameter in the input layer of input neurons and the gain values ​​at respective node values ​​within the hidden layer of the neural network from the integrated adaptive neural network.

[0029] The neural network(s) may further be configured to receive at least one environmental variable as a feedforward term.

[0030] The environmental variable may include at least one of the following variables commonly found in a vehicle motion control system: - the speed of the vehicle for most of the vehicle movement control, - Road / traffic measurements such as curves, direction of travel, speed limits, etc. - Vehicle acceleration / reaction force - Motor speed / angle for current / torque control - Steering torque or angle - Vehicle indicator status, e.g. turning indicated or turning not indicated

[0031] During use of the control system, the weights of the neural network(s) can be updated by a gradient descent backpropagation method each time a new set of input values ​​is fed to the neural network, and the updated weights are used together with the input values ​​to update the neurons.

[0032] According to a fourth aspect, there is provided a vehicle comprising the motion control system according to any one of the preceding claims.

[0033] It will be understood by a person skilled in the art that a feature described in relation to one of the above aspects may also be applied to any other aspect, unless they are mutually exclusive.

[0034] Embodiments of the invention will now be described by way of example only and with reference to the accompanying drawings: Fig. Figure 1 shows a schematic representation of a lateral movement control system incorporated in an autonomous vehicle Fig. 2 shows a schematic representation of a trajectory control of the Fig. 1 shown system for controlling lateral movement; Fig. Figure 3 illustrates the curve, lane position, and heading angle parameters of the autonomous vehicle’s trajectory; Fig. 4 shows a schematic diagram of a general adaptive PID neural network controller provided in the trajectory controller of the motion control system; Fig. Figure 5 is a diagram of the neuron connections of the neural network of the adaptive PID neural network control; Fig. Figure 6 shows a schematic representation of another example of a trajectory control of the lateral movement control system of Fig. 1, in which a curve control is connected in parallel with a lane position control and a heading angle control; Fig. Figure 7 shows a schematic representation of another example of a trajectory control of the lateral movement control system of Fig. 1, in which the curve control is connected in series with the lane position control and the heading angle control; Fig. Figure 8 shows a schematic representation of another example of a trajectory control of the lateral movement control system of Fig. 1, in which a curve control is connected in parallel with an integrated adaptive neural network control for the lane position and the heading angle; and Fig. Figure 9 shows a schematic representation of another example of a trajectory control of the lateral movement control system of Fig. 1, in which a curve control is connected in series with the integrated adaptive neural network control for the lane position and the heading angle.

[0035] Fig. Figure 1 shows a schematic representation of a lateral movement control system 100 installed in an autonomous vehicle 101. The vehicle may be a motor vehicle such as a car, truck, van, or the like. The vehicle 101 may be a fully autonomous vehicle, controlled entirely without user input to control its movement, or the vehicle 101 may be a semi-autonomous vehicle, in which at least the steering of the vehicle is controlled automatically.

[0036] The motion control system 100 generally includes a trajectory planner 102, a trajectory controller 104, and a steering system 106. The trajectory planner 102 is configured to generate a target trajectory for the vehicle 101. The target trajectory may represent a route or a portion of a route to a planned destination of a vehicle's journey. The target trajectory may be determined by the trajectory planner 102 based on sensing information provided by one or more sensors 110. The sensors 110 may include a video camera configured to process images of the surroundings of the vehicle 101. Based on such images, the trajectory planner 102 may calculate an appropriate trajectory for the vehicle 101 so that it can travel autonomously on a road and respond to other vehicles or road conditions.The sensors 110 may be any other suitable sensors configured to provide information from which the vehicle trajectory can be calculated, e.g., vehicle speed, acceleration, etc. The trajectory planner 102 may also have access to navigation information (e.g., from a satellite-based navigation system) to determine the target trajectory.

[0037] The motion control system 100 further includes a trajectory controller 104 configured to receive the target trajectory from the trajectory planner 102 and to provide an output signal including a steering control signal (δ s_tgt ). The steering control signal is generated by the trajectory controller 104 to control the lateral trajectory of the vehicle 101 so that it follows the target trajectory. The trajectory controller 104 outputs the steering control signal δ s_tgtto the and a steering actuation system. The steering system may, for example, include an electric power steering (EPS) or a steer-by-wire (SbW) steering system. The trajectory controller 104 may pass the steering control signal to the steering control system, which provides an output signal to actuate a motor controller of the vehicle steering system 106. The vehicle steering system may include an electric steering system in which an electric motor applies torque to a portion of the vehicle's steering system (e.g., a rack) to steer the wheels of the vehicle 101 or adjust their angle.

[0038] An embodiment of the trajectory control 104 for controlling lateral movement of the vehicle is shown in Fig. 2 in more detail. The trajectory controller 104 is configured to receive a target trajectory consisting of three target trajectory parameters—a target curve path (CV), a target track position (LP), and a target heading angle (CA). These parameters are described with reference to Fig. 3 defined.

[0039] Fig. Figure 3 shows the vehicle while driving on a lane of a road or highway defined by a lane marking labelled 111. The curve of the trajectory refers to the curve of the road the vehicle follows, as shown in Fig. 3. The curve shape can be defined as the inverse of the radius of a circle that best approximates the curve shape of the trajectory at any point along its length.

[0040] Lane position refers to the extent to which the vehicle is offset from the centerline of the lane or roadway it is traveling on. As described in Fig. As can be seen in Figure 3, the lane position is determined by a lateral distance between the center line of the lane (in Fig. 3 marked with X) and the center line of vehicle 101 (in Fig. 3 marked with Y).

[0041] The heading angle refers to the direction the vehicle is traveling on the road. The heading angle can be defined as the angle between the vehicle's intended direction of travel and a reference direction, such as north.

[0042] As in Fig. 2, the trajectory controller 104 comprises three adaptive neural network (NN) controllers, each of which is connected to one of the Fig. 3. The trajectory controller 104 includes an adaptive curve NN controller 110, an adaptive lane position NN controller 112, and an adaptive heading angle NN controller 114. The curve controller 110 is configured to control the curve of the vehicle's trajectory. It receives the target curve from the trajectory planner and provides as output a curve control signal (δ s_cv ). The lane position controller 112 is configured to control the lane position of the vehicle's trajectory. It receives the target lane position from the trajectory planner and provides as output a lane position control signal (δ s_po ). The heading angle controller 114 is configured to control the heading angle of the vehicle's trajectory. It receives the target heading angle from the trajectory planner and provides as output a heading angle control signal (δ s_ca). The trajectory control 104 is configured to calculate (through a summing unit 115) the sum of the control signals (δ s_cv , δ s_po , δ s_ca ) to determine the steering control signal δ s_tgt which is supplied to the steering system 106 as described above.

[0043] Each of the Fig. The adaptive NN controller shown in Figure 2 is configured to minimize an error value that indicates the difference between the actual trajectory of the vehicle and the input target trajectory parameter. By minimizing the error value, the vehicle is controlled to maintain a trajectory that corresponds to a minimal deviation from the target trajectory generated by the trajectory planner 102. Each of the Fig. The adaptive NN controllers 110, 112, 114 shown in Figure 2 (and any other adaptive or non-adaptive controller described herein) receive a general measurement input y. This reflects the current measured value of the trajectory parameter(s) used as input to the feedback loop implemented by the controller. The input y may include measured values ​​of lane position, heading angle, and / or curve progression.

[0044] Each of the above adaptive NN controllers 110, 112, 114 is a PID controller configured to control its I and / or D gain term. Thus, "PID controller" may refer to a P, I, D, PD, PI, or PID controller. In the Fig. In the embodiment shown in Figure 2, the turn control 110 and the heading angle control 114 are PD controls that use the P and D gain terms. The lane position control is a PID control that uses all three gain terms. In other embodiments, the lane position control 112 may not use the D gain term, but may use the P and I gain terms, since the D gain term may be noisy during position measurement and processing. Each of the presently described controls 110, 112, 114 may be a PID control that uses a P and / or an I and / or a D gain term.

[0045] The adaptive NN controllers each comprise a control circuit 104a comprising a PID controller 116 connected to a corresponding neural network 118, as shown in Fig. 4. In the presently described example, the trajectory controller 104 includes a turn control configured to receive an input dependent on the target turn, the turn control coupled to a turn neural network configured to provide P, I, and D gain values ​​as needed. The trajectory controller 104 further includes a lane position controller configured to receive an input dependent on the target lane position, the lane position controller coupled to a lane position neural network. The trajectory controller 104 further includes a heading angle controller configured to receive an input dependent on the target heading angle, the heading angle controller coupled to a lane position neural network.

[0046] Each of the PID controllers 116 generates an output signal u c (e.g. the control signals δ s_cv , δ s_po , δ s_ca ) as the sum of one or more different terms - a proportional term P, an integral term I and a differential term D. Each of these terms is obtained by multiplying an error signal e y with a corresponding gain term K p , K i and K d The error signal is the difference between a target value of a described target curve, the target lane position, or the target heading angle and an actual (current) value of the respective trajectory parameter.

[0047] The gain terms of the PID controller 116 are each calculated by the neural network 118. In the present example, each of the P, I, and D gain terms required for the respective controller is calculated by the neural network, but in other embodiments, one or more of them may be calculated in this way. The gain terms may be determined and updated by the neural network during use of the control system 100 and stored in a memory of the control circuit 115 so that they are available to the controller 116. The P, I, and D gain terms are therefore recalculated and updated at regular intervals during operation of the motion controller and can therefore be continuously adjusted over time to optimize the performance of the PID controller.This is an advantage over using fixed, predetermined values ​​for the PID gain terms, which cannot adapt in the same way.

[0048] The neural network 118 receives as inputs the error signal e y (in Fig. 5 marked with “e”, as described later), the control signal output u c and one or more environment variables U. Each of these variables is provided as a function of the time step k of the neural network's optimization processing. The signals input to the neural network can be updated at regular intervals, and between each update, the neuron values ​​can be updated in response before updated values ​​are input to the neural network.

[0049] The environment variable provided as input to the neural network may be a variable that specifies a parameter not used in the PID controller's control loop. The environment variable may include at least one of the following variables commonly found in a vehicle motion control system: - the speed of the vehicle, - the lane curve - Road / traffic measurements such as lane curves and speed limits, etc. - Vehicle acceleration / reaction force - Motor speed / angle for current / torque control - Steering torque or angle - Vehicle indicator status, e.g. turn indicated or turn not indicated.

[0050] Each of the environmental variables will change during vehicle operation, and providing at least one additional variable at the input point of the neural network allows for further adaptation to the lateral movement control, so that the error value can be better minimized according to the changing environmental conditions during operation. Those skilled in the art will appreciate that this list is not exhaustive and that other variables may also be used within the scope of the invention. In other embodiments, the environmental parameter may not be used at all.

[0051] The neural network 118 comprises a set of neurons 120, as shown in Fig. 5. The neurons comprise an input layer of neurons 120a, a hidden layer of neurons 120b, and an output layer of neurons 120c. The neurons 120 are arranged in a network of connections, with each connection providing the output of one neuron as input to another neuron. Each connection is assigned a weight representing its relative importance. The propagation function calculates the input for a neuron (activation function) from the outputs of its predecessor neurons and their connections as a weighted sum.

[0052] The neural network of Fig. The embodiment illustrated in Figure 5 is a feedforward network with one hidden layer. The neural network is configured to perform gradient descent backpropagation learning to provide the P-gain, I-gain, and D-gain terms used by the respective PID controller. The neural network in this example is a single hidden layer neural network, where the hidden layer has three neurons, each defining a controller. In input layer 120a, the neural network receives a vector X consisting of the error signal e at time steps k, k-1, k-2, etc., at which the neural network operates, the environmental parameter U, and the PID output control signal u cThe number of previous time steps used by the neural network may depend on the type of neural network used. The output of the neural network is the required PID gains K p , K i , K d and the control signal u c . In some embodiments, the output of the neural network may not include the control signal u c . The control signal u c can be provided as the output of the neural network in the presently described embodiment, so that it can be used directly as a PID control output. The lateral movement control system 100 of the present application can, for example, be part of a larger general motion control system whose other components can use the output of the neural network with or without c can use.

[0053] The neural network of this example is configured to perform a gradient descent backpropagation (GDBP) scheme to determine the linear combination weights between neurons. At each time step (k), the controller is configured to perform the following steps: - Updating the values ​​of the input vector X; - Updating the values ​​of the PID gains K p , K i , K d and the control signal u c based on a neuron point calculation from the input X and the weight gains W; - Updating the learning rate for the GDBP scheme; and - Updating the values ​​of the weights W with the GDBP scheme by combining the input vector X and the PID gain states.

[0054] It will be understood by those skilled in the art that the type of neural network described in the present application is only a preferred example. In other embodiments, other types of neural networks or neural network optimization or processing algorithms may also be used.

[0055] In the Fig. In the example of the trajectory controller 104 shown in Figure 2, an adaptive NN controller is provided for each of the respective parameters of the target trajectory. However, in other embodiments, this may not be the case, so that an adaptive NN controller is provided only for one or more of the target trajectory parameter feedback loops. In such examples, a PID controller similar to that shown in Fig. 4, without the presence of a neural network. Instead, the values ​​of the gain terms used by such a PID controller are predetermined and stored in memory so that they are accessible to the controller. The gain terms, in this case, remain constant during operation of the motion control system 100. The gain terms may additionally or alternatively be dependent on one or more vehicle motion variables. The gain terms may, for example, be obtained from a lookup table with predetermined values ​​based on a vehicle motion variable (e.g., vehicle speed) or based on a predetermined functional relationship between the gain terms and the vehicle speed.

[0056] Fig. 6 shows an embodiment of the trajectory control 104 in which only the lane position and the heading angle are provided with an adaptive neural network control, while the curve control is carried out with a non-adaptive control without a neural network, which provides periodically updated gain values.

[0057] In the Fig. 6, the trajectory control 104 includes an adaptive lane position NN control 112, which operates similarly to Fig. 2. The lane position controller 112 performs an adaptive control with an error signal determined by the difference between the target lane position P supplied by the trajectory planner 102 LACand the current or actual value y of the lane position. The gain terms used by the lane position controller 112 are provided by a lane position neural network as described above. The lane position controller provides an output δ fb_p (ie a The trajectory control 104 from Fig. 6 further includes an adaptive heading angle NN control 114. The heading angle control 114 is configured to receive as input a sum of the output of the lane position control 112 (δ fb_p ) and a target course angle parameter CA LACThe sum is calculated by a first computing unit 115a. The heading angle controller 114 then performs adaptive control with an error signal determined by the difference between the target heading angle provided by the trajectory planner 102 and a corresponding current or actual value y of the heading angle parameter. The gain terms used by the heading angle controller 114 are similarly provided by a heading angle neural network as described above. The heading angle controller 114 then provides an output signal δ fb (ie a lane position control signal).

[0058] The adaptive NN lane position and heading angle controllers 112, 114 therefore operate in series with each other, with the output of the lane position controller 112 being summed with the target heading angle and then provided as input to the heading angle controller 114.

[0059] The trajectory control 104 of the Fig. The embodiment illustrated in Figure 6 further includes a curve controller 110'. The curve controller is not equipped with a neural network, but operates with constant values ​​for the required P, I, and D gains that do not change during use of the motion control system. The P, I, and D gain values ​​can be preset before the first use of the motion control system, but otherwise remain constant during subsequent use. The P, I, and D terms can also be dependent on one or more vehicle motion variables, as described above. For example, they can be determined based on a fixed relationship between one or more motion variables and the gain values ​​defined by a lookup table or a fixed functional relationship. The curve controller receives as input a target curve CVtgt the trajectory and performs control using an error signal input by the difference between the target curve and curve control 110' (in Fig. 7 by the parameter y). The curve control 110' is configured to provide an output signal δ ff (ie an output waveform control signal) is generated.

[0060] To generate a final steering control signal, the trajectory controller includes a second summing unit 115b configured to take the sum of the output of the curve control 110' δ ff and the output of the heading angle control 114 δ fb determined to output steering control signalδ f_tgt to deliver as in Fig. 6. The output steering control signal δ f_tgt can then be provided to the steering system 106 as described above.

[0061] The curve control 110' in the Fig. 6 operates in parallel with the lane position control 112 and the heading angle control 114. This may be preferred because it allows the curve control to be included earlier in the calculation of the control signal output by the trajectory control 104. In an alternative embodiment shown in Fig. As shown in Figure 7, the controls for lane position, heading angle, and curve progression are each connected in series. In the embodiment of Fig. 7, the lane position control 112 and the course angle control 114 remain opposite Fig. 7 unchanged and give a control signal CV tgt_fb However, in this embodiment, the output of the curve control 114 is compared with the target curve CV tgtby a second summing unit 115b, the result being provided to the curve controller 110'. The curve controller 110' is again a non-adaptive PID controller without a neural network, similar to the embodiment in Fig. 7. The output system of the curve control 110' provides the final steering control signal δ f_tgt from the trajectory control 104.

[0062] The Fig. 8 and Fig. 9 show further embodiments of a trajectory control 104 which is part of the Fig. 1 shown motion control system 100. In the embodiments of the Fig. 8 and Fig. 9 is an integrated adaptive PID neural network (NN) controller 124 for the controller 124 is used instead of the Fig. 6 and Fig. 7 are used for the lane position and heading angle. The integrated adaptive PID NN controller 124 includes an integrated PID controller that is similar to the Fig. 4, and a neural network corresponding to the one shown in Fig. 4 to provide the gain terms P, I and D. In both embodiments of the Fig. 8 and Fig. 9, the trajectory controller 104 further includes a non-adaptive curve control 110' that does not have a neural network for providing the gain terms P, I, and D, which are instead fixed and / or dependent on one or more vehicle motion variables, as described above. The curve control 110' may be operated in a similar manner as described above in connection with the Fig. 7 and Fig. 8 described.

[0063] The integrated controller 124 is configured to receive as input the target lane position parameter P LAC from the trajectory planner 102. The integrated controller 124 is configured to perform adaptive control with the lane position error (ΔP) and the heading angle error (ΔCA). The integrated PID controller 124 is configured to receive an input signal that depends on a target lane position parameter of the target trajectory and is configured to minimize an error value indicative of the difference between the actual trajectory of the vehicle and the input signal. The integrated PID controller 124 is further configured to receive an input signal that depends on a target heading angle parameter of the target trajectory and is configured to minimize an error value indicative of the difference between the actual trajectory of the vehicle and the input signal.

[0064] The integrated controller 124 is configured similarly to the adaptive PID-NN lane position controller 112 and its associated neural network 118, except that the D gain term provided by the neural network is used to control the heading angle, while the P and I gain terms provided by the neural network are used to control the lane position. The PID and I gain terms provided by the integrated PID controller 124 are used to control the lane position. This is possible because the heading angle can be associated with the derivative of a vehicle's lane position while driving. Controlling the heading angle is equivalent to controlling the derivative of the lane position, so they can share a PID controller.

[0065] By using an integrated adaptive PID-NN controller, a single controller can be used to control both trajectory parameters. This can improve the efficiency of trajectory control and mean that only a few components are required.

[0066] In the Fig. In the embodiment shown in Figure 8, the output signal δ f_fb the integrated controller 124 with the output signal δ f_ff the curve control 110' in a summing unit 115c. The output of the summing unit provides the total control signal δ f_tgtthe trajectory controller 104, which is provided to the steering system 106 as described above. In this embodiment, the integrated controller 124 and the curve control 110' therefore operate in parallel with each other, with their resulting output signals summed to provide the output of the trajectory controller 104. This is similar to the trajectory control in Fig. and enables curve control to be included in the calculation at an early stage.

[0067] Fig. Figure 9 shows an alternative embodiment in which the integrated controller 124 and the curve controller 112' are connected in series. In this embodiment, the output signal CV tgt_fb with the target curve parameter CV tgtsummed by a summing unit 115d. The resulting signal is provided as an input signal to the curve control 110', the output of the curve control 110' being the total control signal δ f_tgt the trajectory control 104.

[0068] In the in the Fig. In the embodiments shown in Figures 6 to 9, the waveform controller uses fixed P, I, and / or D gain terms (e.g., determined using a fixed relationship between one or more motion variables and the gain values, where the fixed relationship is through a). In other embodiments, the waveform controller of these embodiments may be an adaptive PID neural network controller, where a neural network provides the gain terms.

[0069] Various modifications will be apparent to one skilled in the art without departing from the scope of the claims. The embodiments described above are merely examples. Any feature of any aspect or embodiment of the disclosure may be used separately or in combination with any other feature of the same or any other aspect or embodiment of the disclosure, and the disclosure encompasses any feature or combination of features disclosed herein.

Claims

[1] A system for controlling lateral movement of an autonomous vehicle, comprising: a trajectory planner configured to generate a target trajectory for the vehicle; and a trajectory controller configured to receive the target trajectory from the trajectory planner and generate as an output a steering control signal for controlling the lateral movement of the vehicle; wherein the trajectory controller comprises a control circuit, wherein the control circuit comprises an integrated adaptive PID neural network controller comprising: a PID controller that receives an input signal dependent on a target lane position parameter of the target trajectory and is configured to minimize an error value indicative of the difference between the actual trajectory of the vehicle and the input signal dependent on the target lane position parameter, wherein the PID controller is further configured to receive an input signal dependent on a target heading angle parameter of the target trajectory and is configured to minimize an error value indicative of the difference between the actual trajectory of the vehicle and the input signal dependent on the target heading angle parameter; and a neural network configured to determine the P-gain, I-gain, and D-gain terms used by the PID controller, wherein the integrated adaptive PID neural network controller is configured to use the P and I gain terms provided by the neural network, to control the lane position of the vehicle, and the D gain term provided by the neural network to control the heading angle of the vehicle, whereby the P, I, and D gain terms used by the PID controller are provided by the same neural network. [2] The motion control system of claim 1, wherein the trajectory controller further comprises a curve control configured to perform P and / or I and / or D control of the curve of the trajectory of the vehicle. [3] A motion control system according to claim 2, wherein the curve control uses predetermined and / or vehicle motion variable dependent P-gain, I-gain and / or D-gain terms that do not change during use of the motion control system. [4] A motion control system according to claim 2, wherein the curve control is an adaptive neural network control whose neural network is configured to provide P-gain, I-gain and / or D-gain terms. [5] A motion control system according to claim 2 or claim 3 or claim 4, wherein: the curve control is configured to receive as input a target curve parameter and to generate as output a curve control signal; the trajectory controller is configured to determine the sum of the turn control signal and the output of the integrated adaptive PID neural network controller to provide the steering control signal output for the trajectory controller. [6] A motion control system according to any one of the preceding claims, wherein the neural network comprises an input layer of neurons, at least one hidden layer of neurons and an output layer having at least one output neuron, the neural network comprising a feedforward neural network that receives, in the input layer of input neurons, the error value associated with the input trajectory parameter and the control signal output by the integrated adaptive PID neural network controller and outputs the gain values ​​at corresponding node values ​​within the hidden layer of the neural network. [7] The motion control system of claim 6, wherein the neural network is further configured to receive at least one environmental variable as a feedforward term. [8] A motion control system according to claim 7, wherein the environmental variable comprises at least one of the following variables commonly found in a vehicle motion control system: - the speed of the vehicle for most of the vehicle movement control, - Road / traffic measurements such as curves, direction of travel, speed limits, etc. - Vehicle acceleration / reaction force - Motor speed / angle for current / torque control - Steering torque or angle - Vehicle indicator status, e.g. turning indicated or turning not indicated [9] A motion control system according to any one of claims 6 to 8, wherein, during use of the control system, the weights of the neural network are updated by a gradient descent backpropagation scheme each time a new set of input values ​​is supplied to the neural network, and the updated weights are used together with the input values ​​to update the neurons. [10] An autonomous motor vehicle comprising the motion control system according to any one of the preceding claims.

Images