METHOD AND APPARATUS FOR AUTOMATICAL RULE LEARNING FOR AUTONOMOUS DRIVING

• US 20190317496A1
• Filed: 04/11/2018
• Published: 10/17/2019
• Est. Priority Date: 04/11/2018
• Status: Active Grant

• Current Assignee: GM Global Technology Operations LLC (General Motors Company)
• Original Assignee: GM Global Technology Operations LLC (General Motors Company)
• Inventors: Korchev, Dmitriy V., Bhattacharyya, Rajan, Jammalamadaka, Aruna
• Granted Patent: US 10,860,022 B2
• Time in Patent Office: Days
• CPC Class Codes: B60W 10/10 including control of change-speed gearings
• CPC Class Codes: B60W 10/18 including control of braking systems
• CPC Class Codes: B60W 10/20 including control of steering systems
• CPC Class Codes: B60W 2520/06 Direction of travel
• CPC Class Codes: B60W 2520/10 Longitudinal speed
• CPC Class Codes: B60W 2556/50 of positioning data, e.g. GPS [Global Positioning System] data
• CPC Class Codes: B60W 50/00 Details of control systems for road vehicle drive control not related to the control of a particular sub-unit , e.g. process diagnostic or vehicle driver interfaces
• CPC Class Codes: B60W 60/00 Drive control systems specially adapted for autonomous road vehicles
• CPC Class Codes: G05D 1/0088 characterized by the autonomous decision making process, e.g. artificial intelligence, predefined behaviours using knowledge based models G06N5/00
• CPC Class Codes: G05D 1/0221 involving a learning process

Abstract

The present application generally relates to a method and apparatus for generating an action policy for controlling an autonomous vehicle. In particular, the method is operative to receive an input indicative of a training event, segmenting the driving episode into a plurality of time steps, generate a parse tree in response to each time step, and generate a most probable parse tree from a combination of the generated parse trees.

Claims

1. A method of controlling a vehicle comprising:

determining an autonomous driving episode in response to a sensor signal;

segmenting the autonomous driving episode into a first event and a second event;

segmenting the first event into a first time step and a second time step and the second event into a third time step and a fourth time step;

generating a first association matrix in response to the first time step, a second association matrix in response to the second time step, a third association matrix in response to the third time step and a fourth association matrix in response to the fourth time step;

generating a transition matrix in response to the first association matrix, the second association matrix, the third association matrix and the fourth association matrix;

generating a 3D tensor in response to the transition matrix;

determining a probable time step in response to the 3D tensor;

generating a probable event in response to the probable time step; and

generating a control signal for controlling an autonomous vehicle in response to the probable event.

2. The method of claim 1 wherein the sensor signal is generated in response to at least one of a radar, a lidar, a global positioning services receiver, and a video camera.

3. The method of claim 1 wherein probable event is stored in a memory until a similar event is encountered by the autonomous vehicle.

4. The method of claim 1 wherein a first probable parse tree is generated in response to the first time step and a second probable parse tree is generated in response to the second time step.

5. The method of claim 1 wherein an optimal parse tree is generated in response to the first time step and the second time step.

6. The method of claim 1 wherein the transition matrix is used to determine an optimal parse tree for a current time step.

7. The method of claim 1 wherein the control signal is generated in response to a second control signal and an archived episode stored in an episodic memory.

8. The method of claim 1 wherein the autonomous driving episode is matched to an archived episode to generate a plurality of predicted events.

9. The method of claim 8 the control signal is generated in response to the plurality of predicted events.

10. The method of claim 1 wherein the probable event is determined by a cognitive module and the control signal is generated by a motion controller.

11. An apparatus comprising:

a sensor for sensing an environment proximate to an autonomous vehicle and generating a sensor signal;

a perception module for detecting an agent within the environment in response to the sensor signal;

an agent motion predictor for predicting a motion of the agent within the environment in response to the detection of the agent;

a cognitive module for determining a driving episode in response to the predicted motion of the agent, for segmenting the driving episode into a first time step and a second time step, for receiving a first driving data indicating a first action in response to the first time step and a second driving data indicating a second action in response to the second driving data, for generating a rule in response to the first action and the second action for generating a control signal in response to the rule; and

a motion control module for receiving the control signal and controlling an autonomous vehicle in response to the control signal.

12. The apparatus of claim 11 wherein the sensor is at least one of a lidar, radar, and camera.

13. The apparatus of claim 11 wherein the cognitive module is operative to use a current location of the agent and the predicted motion of the agent to generate the control signal.

14. The apparatus of claim 11 wherein the cognitive module has a semantic memory and wherein the semantic memory is operative to generate a new event in response to the first time step and the second time step.

15. The apparatus of claim 11 wherein the cognitive module is operative to generate the rule in response to a temporal grammar and store the rule in a semantic memory for use in generating the control signal.

16. A method of controlling a vehicle comprising;

receiving an input indicative of a training event;

segmenting the driving episode into a first time step and a second time;

receiving a first driving data indicating a first action in response to the first time step and a second driving data indicating a second action in response to the second driving data;

generating a rule in response to the first action and the second action; and

controlling an autonomous vehicle in response to a driving event and the rule.

17. The method of claim 16 wherein input is received from a sensor and is indicative of the driving event.

18. The method of claim 16 wherein the rule is stored in a semantic memory and wherein the rule is retrieved in response to a detected driving scenario.

19. The method of claim 16 wherein a semantic memory is used to generate the rule.

20. The method of claim 16 wherein the rule is generated in response to a temporal grammar and stored in a semantic memory, and wherein the rule is use for generating a control signal for controlling the autonomous vehicle.