System for tuning a trajectory tracking controller for an automotive vehicle

Patent metadata

Field	Value
Publication number	US-12012120-B2
Application number	US-202217874451-A
Country	US
Kind code	B2
Filing date	Jul 27, 2022
Priority date	Jul 27, 2022
Publication date	Jun 18, 2024
Grant date	Jun 18, 2024

Abstract

A system for tuning a trajectory tracking controller for a vehicle includes a trajectory planner configured to generate the planned trajectory and to output one or more planned trajectory components representative of the planned trajectory, a model predictive controller including an internal model and an optimizer, and a tuning neural network configured to receive the one or more planned trajectory components and one or more measured trajectory components and to produce weights for a cost function. The internal model is configured to receive a predicted control input from the optimizer and the one or more measured trajectory components and to produce a predicted output. The optimizer utilizes a cost function and is configured to receive the weights for the cost function and a predicted error and to produce the predicted control input, wherein the predicted error is a selected one of the planned trajectory components minus the predicted output.

First claim

What is claimed is:

1. A system for an automotive vehicle, comprising:
a trajectory planner, comprising a chipset, configured to generate a planned trajectory and to output one or more planned trajectory components representative of the planned trajectory;
a model predictive controller (MPC), comprising a chipset, configured to:
receive a predicted control input and one or more measured trajectory components from the automotive vehicle,
produce a predicted output,
utilize a cost function to receive a plurality of weights for the cost function and a predicted error to produce the predicted control input,
wherein the predicted error is a sum of a selected one of the planned trajectory components minus the predicted output; and
a tuning neural network, comprising a chipset, configured to receive the one or more planned trajectory components and the one or more measured trajectory components and to produce the plurality of weights for the cost function.

2. The system of claim 1 , wherein the predicted control input produced by the MPC is one or more of a steering command, a throttle command and a brake command for the automotive vehicle.

3. The system of claim 1 , wherein the plurality of weights for the cost function are represented by a weights matrix produced by the tuning neural network.

4. The system of claim 1 , wherein the automotive vehicle has a current position and a current speed at a current time and the trajectory planner generates a planned next position and a planned next speed for the automotive vehicle for a next time which is a predetermined time step after the current time, and wherein the one or more planned trajectory components include one or more of:
a forward component of the planned next position;
a lateral component of the planned next position; and
a magnitude of the planned next speed.

5. The system of claim 1 , further comprising:
a critic neural network, comprising a chipset, configured to receive the one or more planned trajectory components and the one or more measured trajectory components;
wherein the tuning neural network and the critic neural network cooperate with each other to tune the plurality of weights utilizing a reinforcement learning optimization algorithm seeking to minimize the cost function.

6. The system of claim 5 , wherein the reinforcement learning optimization algorithm is one of:
a Proximal Policy Optimization algorithm;
a Soft Actor-Critic algorithm;
a Deep Deterministic Policy Gradient algorithm; and
a Twin-Delayed Deep Deterministic Policy Gradient algorithm.

7. The system of claim 5 , wherein the tuning neural network and the critic neural network cooperate with each other to tune the plurality of weights during a training phase.

8. The system of claim 1 , further comprising:
a multi-dimensional look-up table of stored weights operatively connected with the MPC.

9. The system of claim 8 , wherein the MPC retrieves the stored weights from the multi-dimensional look-up table during a deployment phase of an operation of the automotive vehicle.

10. A system for an automotive vehicle, comprising:
a trajectory planner, comprising a chipset, configured to generate a planned trajectory and to output one or more planned trajectory components representative of the planned trajectory;
a model predictive controller (MPC), comprising a chipset, configured to:
receive a predicted control input and one or more measured trajectory components from the automotive vehicle,
produce a predicted output,
utilize a cost function to receive a plurality of weights for the cost function and a predicted error to produce the predicted control input,
wherein the predicted error is a sum of a selected one of the planned trajectory components minus the predicted output; and
a tuning neural network, comprising a chipset, configured to receive the one or more planned trajectory components and the one or more measured trajectory components and to produce the plurality of weights for the cost function; and
a critic neural network, comprising a chipset, configured to receive the one or more planned trajectory components and the one or more measured trajectory components,
wherein the tuning neural network and the critic neural network cooperate with each other to tune the plurality of weights utilizing a reinforcement learning optimization algorithm seeking to minimize the cost function.

11. The system of claim 10 , wherein the predicted control input produced by the MPC is one or more of a steering command, a throttle command and a brake command for the automotive vehicle.

12. The system of claim 10 , wherein the plurality of weights for the cost function are represented by a weights matrix produced by the tuning neural network.

13. The system of claim 10 , wherein the automotive vehicle has a current position and a current speed at a current time and the trajectory planner generates a planned next position and a planned next speed for the automotive vehicle for a next time which is a predetermined time step after the current time, and wherein the one or more planned trajectory components include one or more of:
a forward component of the planned next position;
a lateral component of the planned next position; and
a magnitude of the planned next speed.

14. The system of claim 10 , wherein the reinforcement learning optimization algorithm is one of:
a Proximal Policy Optimization algorithm;
a Soft Actor-Critic algorithm;
a Deep Deterministic Policy Gradient algorithm; and
a Twin-Delayed Deep Deterministic Policy Gradient algorithm.

15. The system of claim 10 , wherein the tuning neural network and the critic neural network cooperate with each other to tune the plurality of weights during a training phase.

16. The system of claim 10 , further comprising:
a multi-dimensional look-up table of stored weights operatively connected with the MPC.

17. The system of claim 16 , wherein the MPC retrieves the stored weights from the multi-dimensional look-up table during a deployment phase of an operation of the automotive vehicle.

18. A trajectory tracking and tuning system for an automotive vehicle, comprising:
a trajectory planner, comprising a chipset, configured to generate a planned trajectory and to output one or more planned trajectory components representative of the planned trajectory;
a model predictive controller (MPC), comprising a chipset, configured to:
receive a predicted control input and one or more measured trajectory components from the automotive vehicle,
produce a predicted output,
utilize a cost function to receive a plurality of weights for the cost function and a predicted error to produce the predicted control input, wherein the predicted error is a sum of a selected one of the planned trajectory components minus the predicted output, and wherein the predicted control input produced by the MPC is one or more of a steering command, a throttle command and a brake command for the automotive vehicle;
a tuning neural network, comprising a chipset, configured to receive the one or more planned trajectory components and the one or more measured trajectory components and to produce the weights for the cost function; and
a critic neural network, comprising a chipset, configured to receive the one or more planned trajectory components and the one or more measured trajectory components, wherein the tuning neural network and the critic neural network cooperate with each other to tune the plurality of weights utilizing a reinforcement learning optimization algorithm seeking to mini

Assignees

• Gm Global Tech Operations Llc

Inventors

• Suissa Avshalom
• Suplin Vladimir

Classifications

• B60W50/0097

  Predicting future conditions · CPC title

• B60W2050/0022

  Gains, weighting coefficients or weighting functions · CPC title

• G06N3/08

  Learning methods · CPC title

• B60W2050/0028

  Mathematical models, e.g. for simulation · CPC title

• B60W30/143

  Speed control (B60W30/16 takes precedence) · CPC title