Determining respective impacts of agents

Patent metadata

Field	Value
Publication number	US-11340622-B2
Application number	US-201916557938-A
Country	US
Kind code	B2
Filing date	Aug 30, 2019
Priority date	Aug 30, 2019
Publication date	May 24, 2022
Grant date	May 24, 2022

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining respective importance scores for a plurality of agents in a vicinity of an autonomous vehicle navigating through an environment. The respective importance scores characterize a relative impact of each agent on planned trajectories generated by a planning subsystem of the autonomous vehicle. In one aspect, a method comprises providing different states of an environment as input to the planning subsystem and obtaining as output from the planning subsystem corresponding planned trajectories. Importance scores for the one or more agents that are in one state but not in the other are determined based on a measure of difference between the planned trajectories.

First claim

What is claimed is:

1. A computed-implemented method comprising:
generating training data for an importance scoring machine learning model configured to process an input comprising data describing a plurality of agents that are present in a scene of an environment and to generate one or more outputs that define a respective importance score for each of the plurality of agents, wherein generating the training data comprises:
obtaining first state data characterizing a first state of the environment with a first set of the plurality of agents being present in the scene of the environment with the autonomous vehicle at a particular time point;
providing a first planning subsystem input comprising the first state data to the planning subsystem of the autonomous vehicle;
in response to providing the first planning subsystem input obtaining from the planning subsystem a first planning subsystem output comprising data that defines a first planned trajectory to be followed by the autonomous vehicle after the particular time point;
obtaining second state data characterizing a second state of the environment with a second set of the plurality of agents being present in the scene of the environment with the autonomous vehicle at the particular time point, wherein the second set of the plurality of agents includes (i) all of the agents that are in the first set of the plurality of agents and (ii) one or more agents that are not in the first set of the plurality of agents;
providing a second planning subsystem input comprising the second state data to the planning subsystem;
in response to providing the second planning subsystem input, obtaining from the planning subsystem a second planning subsystem output comprising data that defines a second planned trajectory to be followed by the autonomous vehicle after the particular time point;
determining a measure of a difference between (i) the first planned trajectory to be followed by the autonomous vehicle after the particular time point and (ii) the second planned trajectory to be followed by the autonomous vehicle after the particular time point;
determining, based at least in part on the determined measure of difference, a respective importance score for each of the one or more agents that are in the second set of the plurality of agents but not in the first set of the plurality of agents, wherein the respective importance score characterizes, for each of the one or more agents that are in the second set of the plurality of agents but not in the first set of the plurality of agents, a relative impact of the agent on the first planned trajectory generated by the planning subsystem of the autonomous vehicle; and
generating the training data that includes (i) data describing the one or more agents that are in the second set of the plurality of agents but not in the first set of the plurality of agents, and (ii) for each of the one or more agents that are the second set of the plurality of agents but not in the first set of the plurality of agents, a label defining the respective importance score for the agent as a ground truth importance score of the agent; and

training the importance scoring machine learning model on the training data.

2. The method of claim 1 , wherein generating the training data further comprises:
obtaining third state data characterizing a third state of the environment with a third set of the plurality of agents being present in the scene of the environment with the autonomous vehicle, wherein the third set of the plurality of agents includes all of the agents that are in the first set of the plurality of agents less one or more agents;
providing a third planning subsystem input comprising the third state data to the planning subsystem;
obtaining from the planning subsystem a third planning subsystem output comprising data that defines a third planned trajectory to be followed by the autonomous vehicle after the particular time point;
determining a measure of a difference between the first and third planned trajectories; and
determining, based at least on the determined measure of difference, a respective importance score for each of the one or more agents that are in the first set of the plurality of agents but not in the third set of the plurality of agents.

3. The method of claim 1 , wherein the first and second states of the environment characterize the scene of the environment except with different numbers of agents that are present in the scene of the environment.

4. The method of claim 3 , wherein the scene of the environment is a scene of a simulated environment.

5. The method of claim 3 , wherein the scene of the environment is a scene of a real world environment that is perceived by one or more sensors onboard the autonomous vehicle.

6. The method of claim 1 , wherein determining the measure of the difference between the first and second planned trajectories comprises:
determining a Euclidean distance between the first and second planned trajectories.

7. The method of claim 1 , wherein the planning subsystem of the autonomous vehicle is a cost-based planning subsystem, and wherein determining the measure of the difference between the first and second planned trajectories comprises:
determining a difference between respective costs of the first and second planned trajectories that are both generated by the cost-based planning subsystem.

8. The method of claim 1 , wherein determining the measure of the difference between the first and second planned trajectories comprises:
determining respective differences between corresponding geometries and speeds of the first and second planned trajectories.

9. A system comprising:
one or more computers; and
one or more storage devices storing instructions that, when executed by the one or more computers, cause the one or more computers to perform operations comprising:
generating training data for an importance scoring machine learning model configured to process an input comprising data describing a plurality of agents that are present in a scene of an environment and to generate one or more outputs that define a respective importance score for each of the plurality of agents, wherein generating the training data comprises:
obtaining first state data characterizing a first state of the environment with a first set of the plurality of agents being present in the scene of the environment with the autonomous vehicle at a particular time point;
providing a first planning subsystem input comprising the first state data to the planning subsystem of the autonomous vehicle;
in response to providing the first planning subsystem input, obtaining from the planning subsystem a first planning subsystem output comprising data that defines a first planned trajectory to be followed by the autonomous vehicle after the particular time point;
obtaining second state data characterizing a second state of the environment with a second set of the plurality of agents being present in the scene of the environment with the autonomous vehicle at the particular time point, wherein the second set of the plurality of agents includes (i) all of the agents that are in the first set of the plurality of agents and (ii) one or more agents that are not in the first set of the plurality of agents;
providing second planning subsystem input comprising the second state data to the planning subsystem;
in response to providing the second planning subsystem input obtaining from the planning subsystem a second planning subsystem output comprising data that defines a second planned trajectory to be followed by the autonomous vehicle after the particular time point;
determining a measure of a difference between (i) the first planned trajectory to be f

Assignees

• Waymo Llc

Inventors

• Refaat Khaled
• Ding Kai
• Ross Stephane

Classifications

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title

• G06N3/08

  Learning methods · CPC title

• G05D1/0221

  involving a learning process · CPC title

• G01C21/3446

  Details of route searching algorithms, e.g. Dijkstra, A*, arc-flags or using precalculated routes · CPC title

• G05D1/0219 Primary

  ensuring the processing of the whole working surface · CPC title