Systems and methods for updating the parameters of a model predictive controller with learned controls parameters generated using simulations and machine learning

Patent metadata

Field	Value
Publication number	US-12459497-B2
Application number	US-202117165822-A
Country	US
Kind code	B2
Filing date	Feb 2, 2021
Priority date	Feb 2, 2021
Publication date	Nov 4, 2025
Grant date	Nov 4, 2025

Abstract

A computer implemented method for determining optimal values for controls parameters for a model predictive controller for controlling a vehicle can receive from a data store or a graphical user interface, ranges for one or more operational parameters. The computer implemented method can determine optimum values for controls parameters by simulating a vehicle operation across the ranges of the one or more operational parameters by solving a vehicle control problem and determining an output of the vehicle control problem based on a result for the simulated vehicle operation. A vehicle can include a processing component configured to adjust a control input for an actuator of the vehicle according to a control algorithm and based on the optimum values of the controls parameter as determined by the computer implemented method.

First claim

What is claimed is:

1 . A computer implemented method for determining and utilizing optimal real time operational parameters to adjust a control input for an actuator of a vehicle comprising a model predictive controller for controlling the vehicle, the method comprising:
receiving, at a hardware processor, from a data store or a graphical user interface, a range for one or more operational parameters associated with the vehicle, the one or more operational parameters comprising one or more controls parameters; determining, by a trained machine learning vehicle performance circuit, optimum real time values for a controls parameter of the one or more controls parameters, wherein the trained machine learning vehicle performance circuit is trained by:
simulating, by a vehicle control circuit, a vehicle operation across the range of the one or more operational parameters by solving a model predictive control problem over a prediction horizon and across one or more real time steps;
determining, by the trained machine learning vehicle performance circuit, an output of the model predictive control problem, the output including one or more of the optimal real time operational parameters based on a result for the simulated vehicle operation;
updating the trained machine learning vehicle performance circuit based on one or more training sets updated with the optimum real time values for training the trained machine learning vehicle performance circuit; and
re-training the trained machine learning vehicle performance circuit with the one or more updated training sets; and
adjusting the control input for the actuator of the vehicle according to the one or more optimal real time operational parameters to control operation of the actuator.

2 . The method of claim 1 , wherein the model predictive control problem comprises the model predictive controller, and wherein the controls parameter comprises one or more of a gain Q on a vehicle state, a gain R on an input to the model predictive controller, or a gain W on a slack to the vehicle state.

3 . The method of claim 1 , wherein the one or more operational parameters comprise another controls parameter, the another controls parameter comprising one or more of a gain Q on a vehicle state, a gain R on an input to the model predictive controller, or a gain W on a slack to the vehicle state.

4 . The method of claim 1 , wherein the one or more operational parameters further comprise a vehicle parameter, and a gain Q on a vehicle state, a gain R on an input, or a gain W on a slack for the controls parameter are gains based on the vehicle parameter.

5 . The method of claim 4 , wherein the vehicle parameter is selected from a group consisting of a distance A from a center of gravity (CG) of the vehicle to a front axle of the vehicle, a distance B from the center of gravity (CG) to a rear axle of the vehicle, a distance L from a center of the front axle to a center of the rear axle of the vehicle, a tire distance from the CG to the rear axle of the vehicle, a vehicle speed V x , a vehicle yaw rate r, vehicle sideslip angle β, front steering angle δ, front and rear lateral tire forces F yf and F yf , a vehicle mass m, a yaw inertia I zz , a height h of the vehicle's CG, a wheel radius R, a cornering stiffness C, a front axle cornering stiffness C af , or a rear axle cornering stiffness C ar .

6 . The method of claim 1 , wherein the vehicle operation is simulated across a range of one or more operational parameters comprising a range for the controls parameter and one or more other operational parameters.

7 . The method of claim 1 , wherein the vehicle operation comprises at least one of a path tracking, corridor keeping, stabilization, or collision avoidance maneuver, and optimum real time values for the one or more operational parameters correspond to the values of the parameters which were determined by the trained machine learning vehicle performance circuit to allow for the vehicle to at least one of:
(i) track the path in the path tracking maneuver, (ii) remain within the corridor in the corridor keeping maneuver; (iii) stabilize the vehicle in the stabilization maneuver, or (iv) maneuver to avoid vehicle collision in the collision avoidance maneuver.

8 . The method of claim 1 , further comprising:
generating a training set of optimum one or more controls parameters, wherein the training set of optimum one or more controls parameters is configured to be used as an initial parameter set for the model predictive controller operating on the vehicle.

9 . The method of claim 1 , wherein the vehicle operation comprises at least one of a path tracking, stabilization, or collision avoidance maneuver, and wherein the trained machine learning vehicle performance circuit is trained by:
simulating, by the vehicle control circuit, the vehicle operation across a full range of the one or more operational parameters by solving the model predictive control problem for at least one of the path tracking, stabilization, or collision avoidance maneuver; and determining the optimum real time value for a first parameter based on outcome of the path tracking, stabilization, or collision avoidance maneuver during simulation.

10 . The method of claim 1 , wherein the one or more operational parameters comprises a first external parameter; and determining by the trained machine learning vehicle performance circuit, optimum real time values for the controls parameter further comprises:
determining by the trained machine learning vehicle performance circuit, a value for controls parameter based on one or more other parameters.

11 . The method of claim 10 wherein the first external parameter is selected from a group consisting of:
a friction coefficient between at least one tire and a road,
a gravitational constant,
a road surface roughness,
an external humidity,
a wind vector, and
an external temperature.

12 . The method of claim 1 , wherein the one or more operational parameters comprise another controls parameter; and wherein the model predictive control problem controls the control input based on:
a current vehicle state; predicted boundaries for values of the one or more operational parameters; and a future vehicle state determined based on the predicted boundaries for values of the one or more operational parameters.

13 . A vehicle, comprising:
a processing component configured to adjust a control input for an actuator of the vehicle according to a control algorithm comprising a model predictive control problem which controls the control input based on a current vehicle state, predicted boundaries for values of operational parameters associated with the vehicle, and based on a future vehicle state determined based on the predicted boundaries of values for one or more operational parameters; a data store coupled to the processing component, wherein the data store contains a value for a controls parameter for execution of the control algorithm; wherein the processing component is configured to adjust the control input for the actuator based on the value for the controls parameter; wherein the value for the controls parameter was received by the processing component by a vehicle simulation system; wherein the vehicle simulation system generated the value for the controls parameter by simulating a vehicle operation across a range of one or more operational parameters over a prediction horizon and across one or more real time steps, and determined an optimum real time value for the controls parameter based on a result for the simulation of the vehicle operation; wherein the processing component i

Assignees

• Toyota Res Inst Inc

Inventors

• Thompson Michael
• Bobier-Tiu Carrie
• Ahumada Manuel
• Bhargava Arjun
• Balachandran Avinash

Classifications

• G05D1/49

  Control of attitude, i.e. control of roll, pitch or yaw · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N20/00

  Machine learning · CPC title

• B60W2530/20

  Tyre data · CPC title

• B60W30/02

  Control of vehicle driving stability · CPC title