Learning-based dynamic modeling methods for autonomous driving vehicles

Patent metadata

Field	Value
Publication number	US-11199846-B2
Application number	US-201816204941-A
Country	US
Kind code	B2
Filing date	Nov 29, 2018
Priority date	Nov 29, 2018
Publication date	Dec 14, 2021
Grant date	Dec 14, 2021

Abstract

In an embodiment, a learning-based dynamic modeling method is provided for use with an autonomous driving vehicle. A control module in the ADV can generate current states of the ADV and control commands for a first driving cycle, and send the current states and control commands to a dynamic model implemented using a trained neural network model. Based on the current states and the control commands, the dynamic model generates expected future states for a second driving cycle, during which the control module generates actual future states. The ADV compares the expected future states and the actual future states to generate a comparison result, for use in evaluating one or more of a decision module, a planning module and a control module in the ADV.

First claim

What is claimed is:

1. A computer-implemented method of evaluating an autonomous driving system of an autonomous driving vehicle (ADV), comprising:
measuring a plurality of current states of the ADV for a first driving cycle, wherein the plurality of current states of the ADV include a speed, an acceleration, and an angular velocity of the ADV;
predicting, using a dynamic model in the ADV, a plurality of expected future states for a second driving cycle of the ADV based on the plurality of current states for the first driving cycle and a first set of control commands issued for the first driving cycle, wherein the first set of control commands include a throttle command, a brake command, and a steering command, wherein the dynamic model is a trained neural network model, wherein the plurality of current states and the first set of control commands are provided as input to the trained neural network model;
measuring a plurality of actual future states for the second driving cycle to control the ADV in response to the first set of control commands; and
comparing the plurality of expected future states and the plurality of actual future states, wherein a result of the comparison is used to modify one or more autonomous driving modules in the autonomous driving system of the ADV.

2. The method of claim 1 , wherein the trained neural network model is one of a linear regression, a multiplayer perceptron (MLP), or a recurrent neural network (RNN).

3. The method of claim 2 , wherein the trained neural network model is trained using datasets that include real-world data collected by sensors in vehicles.

4. The method of claim 1 , wherein the ADV further includes an inverse dynamic model, which is configured to compute one or more control commands based on the plurality of current states and the plurality of expected future states, wherein the one or more control commands are used for real-time adaptive control of the ADV.

5. The method of claim 1 , wherein the result of the comparison is persisted in a persistent storage for future use.

6. The method of claim 1 , wherein each of the first driving cycle and the second driving cycle is a time interval during which the ADV generates planning and control data, and issues one or more control commands based on the planning and control data.

7. A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, causing the processor to perform operations of evaluating an autonomous driving system of an autonomous driving vehicle (ADV), the operations comprising:
measuring a plurality of current states of the ADV for a first driving cycle, wherein the plurality of current states of the ADV include a speed, an acceleration, and an angular velocity of the ADV;
predicting, using a dynamic model in the ADV, a plurality of expected future states for a second driving cycle of the ADV based on the plurality of current states for the first driving cycle and a first set of control commands issued for the first driving cycle, wherein the first set of control commands include a throttle command, a brake command, and a steering command, wherein the dynamic model is a trained neural network model, wherein the plurality of current states and the first set of control commands are provided as input to the trained neural network model;
measuring a plurality of actual future states for the second driving cycle to control the ADV in response to the first set of control commands; and
comparing the plurality of expected future states and the plurality of actual future states, wherein a result of the comparison is used to modify one or more autonomous driving modules in the autonomous driving system of the ADV.

8. The non-transitory machine-readable medium of claim 7 , wherein the trained neural network model is one of a linear regression, a multiplayer perceptron (MLP), or a recurrent neural network (RNN).

9. The non-transitory machine-readable medium of claim 8 , wherein the trained neural network model is trained using datasets that include real-world data collected by sensors in vehicles.

10. The non-transitory machine-readable medium of claim 7 , wherein the ADV further includes an inverse dynamic model, which is configured to compute one or more control commands based on the plurality of current states and the plurality of expected future states, wherein the one or more control commands are used for real-time adaptive control of the ADV.

11. The non-transitory machine-readable medium of claim 7 , wherein the result of the comparison is persisted in a persistent storage for future use.

12. The non-transitory machine-readable medium of claim 7 , wherein each of the first driving cycle and the second driving cycle is a time interval during which the ADV generates planning and control data, and issues one or more control commands based on the planning and control data.

13. A data processing system, comprising:
a processor; and
a memory coupled to the processor to store instructions, which, when executed by a processor, cause the processor to perform operations of evaluating an autonomous driving system of an autonomous driving vehicle (ADV), the operations comprising:
measuring a plurality of current states of the ADV for a first driving cycle, wherein the plurality of current states of the ADV include a speed, an acceleration, and an angular velocity of the ADV;
predicting, using a dynamic model in the ADV, a plurality of expected future states for a second driving cycle of the ADV based on the plurality of current states for the first driving cycle and a first set of control commands issued for the first driving cycle, wherein the first set of control commands include a throttle command, a brake command, and a steering command, wherein the dynamic model is a trained neural network model, wherein the plurality of current states and the first set of control commands are provided as input to the trained neural network model;
measuring a plurality of actual future states for the second driving cycle to control the ADV in response to the first set of control commands; and
comparing the plurality of expected future states and the plurality of actual future states, wherein a result of the comparison is used to modify one or more autonomous driving modules in the autonomous driving system of the ADV.

14. The system of claim 13 , wherein the trained neural network model is one of a linear regression, a multiplayer perceptron (MLP), or a recurrent neural network (RNN).

15. The system of claim 14 , wherein the trained neural network model is trained using datasets that include real-world data collected by sensors in vehicles.

16. The system of claim 13 , wherein the ADV further includes an inverse dynamic model, which is configured to compute one or more control commands based on the plurality of current states and the plurality of expected future states, wherein the one or more control commands are used for real-time adaptive control of the ADV.

17. The system of claim 13 , wherein the result of the comparison is persisted in a persistent storage for future use, and wherein each of the first driving cycle and the second driving cycle is a time interval during which the ADV generates planning and control data, and issues one or more control commands based on the planning and control data.

Assignees

• Baidu Usa Llc

Inventors

• Luo Qi
• Xu Jiaxuan
• Xu Kecheng
• Xiao Xiangquan
• Yu Siyang
• Miao Jinghao
• Hu Jiangtao

Classifications

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/048

  Activation functions · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/0499

  Feedforward networks · CPC title