Method and system for predicting external agent behavior

Patent metadata

Field	Value
Publication number	US-12545290-B2
Application number	US-202418982386-A
Country	US
Kind code	B2
Filing date	Dec 16, 2024
Priority date	Dec 14, 2023
Publication date	Feb 10, 2026
Grant date	Feb 10, 2026

Abstract

A method for predicting external agent behavior can include: receiving decisioning data, identifying a set of environmental agents (e.g., external agents), and evaluating a set of hypotheses for each environmental agent. A system for predicting external agent behavior can include a computing system and a sensor suite (e.g., onboard an autonomous vehicle), which can function to implement any or all of the processes of the method.

First claim

We claim:

1 . A method, comprising:
collecting data from a set of sensors onboard an autonomous vehicle; identifying an environmental agent in an environment of the autonomous vehicle; generating a tree model based on a set of decisions, the tree model comprising nodes representing hypotheses for the environmental agent, wherein generating the tree model comprises:
for a parent decision in the set of decisions: using a parent decision model associated with the parent decision, determining a probability for each of a set of parent hypotheses for the parent decision based on a first subset of the data; and
for a child decision in the set of decisions: for each parent hypothesis in the set of parent hypotheses: using a child decision model associated with the child decision, determining a probability for each of a set of child hypotheses for the child decision based on a second subset of the data;
determining a set of environmental agent policy options for the environmental agent based on the tree model, wherein each environmental agent policy option in the set of environmental agent policy options is defined by a branch of the tree model comprising a set of hypotheses; performing a set of simulations based on the set of environmental agent policy options and a set of ego agent policy options for the autonomous vehicle, wherein performing each simulation in the set of simulations comprises:
selecting an ego agent policy option from the set of ego agent policy options;
selecting an environmental agent policy option from the set of environmental agent policy options, wherein a probability of selecting the environmental agent policy option is determined based on a probability distribution of the set of environmental agent policy options, the probability distribution determined based on the tree model;
simulating the ego agent policy option and the environmental agent policy option as a joint policy set;
selecting an ego agent policy for the autonomous vehicle based on the set of simulations; and operating the autonomous vehicle according to the selected ego agent policy.

2 . The method of claim 1 , wherein, for each parent hypothesis in the set of parent hypotheses, the probability for each of the set of child hypotheses is further determined based on the probability for the parent hypothesis.

3 . The method of claim 1 , wherein, for each parent hypothesis in the set of parent hypotheses, the child decision model associated with the child decision is selected based on the parent hypothesis.

4 . The method of claim 3 , wherein the parent decision comprises an agent classification, wherein the set of parent hypotheses comprises a vehicle classification and a pedestrian classification, wherein the child decision is associated with a first child decision model for the child decision and a second child decision model for the child decision, wherein the first child decision model for the child decision corresponds to the vehicle classification and the second child decision model for the child decision corresponds to the pedestrian classification.

5 . The method of claim 1 , wherein the set of parent hypotheses for the parent decision are predetermined.

6 . The method of claim 5 , wherein the set of child hypotheses for the child decision are determined based on the data.

7 . The method of claim 1 , wherein the set of decisions comprise at least: an agent classification decision, a route decision, a decision for an environmental agent response to the autonomous vehicle, and an acceleration decision.

8 . The method of claim 1 , wherein a decision in the set of decisions comprises a jaywalking decision.

9 . A method, comprising:
collecting data from a set of sensors onboard an autonomous vehicle; identifying an environmental agent in an environment of the autonomous vehicle; generating a tree model based on a set of decisions, the tree model comprising nodes representing hypotheses for the environmental agent, wherein generating the tree model comprises:
for a parent decision in the set of decisions: using a parent decision model associated with the parent decision, determining a probability for each of a set of parent hypotheses for the parent decision based on a first subset of the data; and
for a child decision in the set of decisions: for each parent hypothesis in the set of parent hypotheses: using a child decision model associated with the child decision, determining a probability for each of a set of child hypotheses for the child decision based on a second subset of the data;
determining a set of environmental agent policy options for the environmental agent based on the tree model, wherein each environmental agent policy option in the set of environmental agent policy options is defined by a branch of the tree model comprising a set of hypotheses; determining a sample of environmental agent policy options selected from the set of environmental agent policy options, wherein the environmental agent policy options in the sample are selected based on a probability distribution of the set of environmental agent policy options, the probability distribution determined based on the tree model, wherein at least one environmental agent policy option is selected multiple times; for each environmental policy agent option in the sample, performing a simulation; selecting an ego agent policy for the autonomous vehicle based on the simulations; and operating the autonomous vehicle according to the selected ego agent policy.

Assignees

• May Mobility Inc

Inventors

• Crossman Jacob
• Borle Kapil
• Saucer Timothy
• Olson Edwin B

Classifications

• G06N20/20 Primary

  Ensemble learning · CPC title

• B60W2554/4029

  Pedestrians · CPC title

• B60W2554/4045

  Intention, e.g. lane change or imminent movement · CPC title

• B60W60/001 Primary

  Planning or execution of driving tasks · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title