Method and system for developing autonomous vehicle training simulations

Patent metadata

Field	Value
Publication number	US-11960292-B2
Application number	US-202117387927-A
Country	US
Kind code	B2
Filing date	Jul 28, 2021
Priority date	Jul 28, 2021
Publication date	Apr 16, 2024
Grant date	Apr 16, 2024

Abstract

Method and systems for generating vehicle motion planning model simulation scenarios are disclosed. The system receives a base simulation scenario with features of a scene through which a vehicle may travel. The system then generates an augmentation element with a simulated behavior for an object in the scene by: (i) accessing a data store in which behavior probabilities are mapped to object types to retrieve a set of behavior probabilities for the object; and (ii) applying a randomization function to the behavior probabilities to select the simulated behavior for the object. The system will add the augmentation element to the base simulation scenario at the interaction zone to yield an augmented simulation scenario. The system will then apply the augmented simulation scenario to an autonomous vehicle motion planning model to train the motion planning model.

First claim

The invention claimed is:

1. A method of generating a vehicle motion planning simulation scenario, the method comprising, by a processor:
receiving, from a first data store containing a plurality of simulation scenarios, a base simulation scenario that includes features of a scene through which a vehicle may travel;
generating an augmentation element that comprises a simulated behavior for an object in the scene by:
accessing a second data store in which behavior probabilities are mapped to object types to retrieve a set of behavior probabilities for the object, and
applying a randomization function to the behavior probabilities to select the simulated behavior for the object;

adding the augmentation element to the base simulation scenario to yield an augmented simulation scenario; and
applying the augmented simulation scenario to an autonomous vehicle motion planning model to train the motion planning model in which the motion planning model:
simulates movement of the vehicle along a planned trajectory,
in response to a perception system of the vehicle detecting the augmentation element, selects a continued trajectory for the vehicle, wherein the continued trajectory is either the planned trajectory or an alternate trajectory, and
causes the vehicle to move along the continued trajectory.

2. The method of claim 1 , wherein the method further comprises using the trained motion planning model to generate a trajectory for the vehicle.

3. The method of claim 1 , wherein:
the simulated behavior for the object at least partially interferes with the planned trajectory of the vehicle; and
the continued trajectory is an alternate trajectory that will keep the vehicle at least a threshold distance away from the object.

4. The method of claim 3 , wherein:
the object comprises a vehicle, a pedestrian, an animal, a bicycle, vegetation, or an unclassifiable object; and
adding the augmentation element to the base simulation scenario comprises at least partially positioning the object in a lane that the planned trajectory will traverse.

5. The method of claim 1 , wherein the simulated behavior is one that a perception system of the vehicle is expected to find ambiguous in that the perception system is expected to assign substantially equal likelihoods to the simulated behavior in response to the simulated behavior corresponding to at least two candidate behaviors which are inconsistent with each other.

6. The method of claim 1 , wherein the simulated behavior is one that a perception system of the vehicle is expected to find ambiguous in response to the simulated behavior being inconsistent with a class of the object.

7. The method of claim 1 , wherein:
the object is a vehicle, pedestrian, bicycle, or other actor; and
the simulated behavior is one that causes movement of the object and which, when taken, will cause the vehicle's motion planning model to react by selecting the alternate trajectory.

8. The method of claim 1 , further comprising:
generating a plurality of augmentation elements across a plurality of segments of an interaction zone in the scene, wherein each of the segments is assigned an element distribution with weights for one or more object classes or one or more behavior types; and
using the element distribution to select the simulated behavior for the object in each of the augmentation elements.

9. A computer program product comprising:
a memory that stores programming instructions that are configured to cause a processor to train a vehicle motion planning model by:
receiving, from a first data store containing a plurality of simulation scenarios, a base simulation scenario that includes features of a scene through which a vehicle may travel,
generating an augmentation element that comprises a simulated behavior for an object in the scene by:
accessing a second data store in which behavior probabilities are mapped to object types to retrieve a set of behavior probabilities for the object; and
applying a randomization function to the behavior probabilities to select the simulated behavior for the object;

adding the augmentation element to the base simulation scenario at the interaction zone to yield an augmented simulation scenario, and
applying the augmented simulation scenario to an autonomous vehicle motion planning model to train the motion planning model by, in the augmented simulation scenario:
simulating movement of the vehicle along a planned trajectory;
by the motion planning model, in response to a perception system of the vehicle detecting the augmentation element, selecting a continued trajectory for the vehicle, wherein the continued trajectory is either the planned trajectory or an alternate trajectory; and
causing the vehicle to move along the continued trajectory.

10. The computer program product of claim 9 , further comprising additional programming instructions that are configured to cause the processor to use the trained motion planning model to generate a trajectory for the vehicle.

11. The computer program product of claim 9 , wherein:
the instructions to apply a randomization function to the behavior probabilities to select the simulated behavior for the object comprise instructions to select the simulated behavior as one that will at least partially interfere with the planned trajectory of the vehicle; and
the instructions to cause the vehicle to move along the continued trajectory comprise instructions to cause the vehicle to move along an alternate trajectory that will keep the vehicle at least a threshold distance away from the object.

12. The computer program product of claim 11 , wherein the instructions to add the augmentation element to the base simulation scenario comprise instructions to at least partially position the object in a lane that the planned trajectory will traverse.

13. The computer program product of claim 9 , wherein the instructions to apply a randomization function to the behavior probabilities to select the simulated behavior for the object comprise one or more of the following:
instructions to select the simulated behavior as one that a perception system of the vehicle is expected to find ambiguous in that the perception system is expected to assign substantially equal likelihoods to the simulated behavior in response to the simulated behavior corresponding to at least two candidate behaviors which are inconsistent with each other;
instructions to select the simulated behavior as one that a perception system of the vehicle is expected to find ambiguous in response to the simulated behavior being inconsistent with a class of the object; or
instructions to select the simulated behavior as one that will cause movement of the object and which, when taken, will cause the vehicle's motion planning model to react by selecting the alternate trajectory.

14. The computer program product of claim 9 , further comprising additional programming instructions that are configured to cause the processor to:
generate a plurality of augmentation elements across a plurality of segments of an interaction zone, wherein each of the segments is assigned an element distribution with weights for one or more object classes or one or more behavior types; and
when applying a randomization function to the behavior probabilities to select the simulated behavior for the object in each of the augmentation elements, use the element distribution to select the simulated behavior for the object.

15. A vehicle motion planning model training system, comprising:
a processor;
a first data store containing a plurality of simulation scenarios; and
a memor

Assignees

• Argo Ai Llc

Inventors

• Nayhouse Michael
• Pacilio Michael
• Ackenhausen Thomas Carl
• Flick Allen Edward

Classifications

• G05D1/0221 Primary

  involving a learning process · CPC title

• B60W30/095

  Predicting travel path or likelihood of collision · CPC title

• B60W30/18163

  Lane change; Overtaking manoeuvres · CPC title

• G05D1/0044

  by providing the operator with a computer generated representation of the environment of the vehicle, e.g. virtual reality, maps (maps used for automatic navigation G05D1/0274; flight directors G01C23/005) · CPC title

• G05D1/0088

  characterized by the autonomous decision making process, e.g. artificial intelligence, predefined behaviours (using knowledge based models G06N5/00) · CPC title