Method and system for feasibility-based operation of an autonomous agent
The method can include: receiving a set of inputs; determining a set of policies based on the set of inputs; determining a set of scores associated with the set of environmental policies; and evaluating the set of policies. Additionally or alternatively, the method can include operating the ego agent according to a selected policy and/or any other processes. The method functions to facilitate scoring of policies based on ‘feasibility’ for agents in an environment. Additionally or alternatively, the method can function to facilitate autonomous operation of a vehicle (e.g., based on policy-feasibility of agents in the environment). Additionally or alternatively, the method can function to facilitate intention estimation for agents in an environment.
1. A method of operation of an autonomous vehicle in an environment, comprising:
determining a set of inputs using a sensor suite of the autonomous vehicle, the set of inputs comprising an environmental agent instance identifier and a state history associated with the environmental agent instance identifier;
based on the set of inputs, determining a set of multiple environmental policies for the environmental agent instance identifier;
for each environmental policy of the set of multiple environmental policies:
determining a historical score by comparing the state history associated with the environmental agent instance identifier to a reference trajectory associated with the environmental policy; and
determining a feasibility score by a forward simulation of the environmental policy, wherein the forward simulation of each environmental policy comprises a closed-loop simulation for a deterministic controller associated with the environmental policy, wherein the feasibility score is determined based on a time-derivative of an accumulation of lateral error between the reference trajectory and the forward simulation; and
aggregating the historical score and the feasibility score for each environmental policy of the set of multiple environmental policies into a respective aggregate score, wherein producing the respective aggregate score comprises multiplying the respective historical score and the respective feasibility score;
determining an ego policy by evaluating a set of ego policies for the autonomous vehicle relative to the set of multiple environmental policies, based on the feasibility scores and the historical scores, wherein the evaluation of the set of ego policies relative to the set of multiple environmental policies is weighted based on the aggregate score of each environmental policy of the set of multiple environmental policies; and
controlling driving of the autonomous vehicle based on the ego policy.
2. The method of claim 1 , wherein evaluating the set of ego policies comprises simulating vehicle policies based on environmental policy scenarios which are sampled from the set of multiple environmental policies based on the aggregate scores.
3. The method of claim 1 , wherein evaluating the set of ego policies comprises performing multiple simulations of each ego policy relative to the set of multiple environmental policies based on the feasibility scores and the historical scores.
4. The method of claim 1 , wherein each feasibility score characterizes a predicted control effort associated with the forward simulation of the environmental policy.
5. The method of claim 4 , wherein the predicted control effort comprises one or more of: a lateral control parameter and longitudinal control parameter.
6. The method of claim 1 , wherein the set of inputs further comprises a classification associated with the environmental agent instance identifier, wherein the set of multiple environmental policies is based on the classification.
7. The method of claim 6 , further comprising:
determining a supplementary score for each environmental policy of the set of multiple environmental policies based on the classification and the state history, wherein the ego policy is determined based further on the supplementary scores.
8. The method of claim 1 , wherein determining each environmental policy of the set of multiple environmental policies for the environmental agent instance identifier comprises extracting the respective reference trajectory from a prior route network.
9. A method for operation of an autonomous vehicle relative to agents in an environment of the autonomous vehicle, comprising:
tracking a set of agents in the environment based on vehicle sensor data, comprising determining a state history of each agent of the set of agents;
determining a set of multiple policy candidates for each agent, wherein each policy candidate comprises a closed-loop, deterministic controller;
determining a respective set of scores for each policy candidate of the set of multiple policy candidates for each agent, comprising:
determining a first score based on a comparison between the policy candidate and the state history of the agent;
determining a second score by a forward simulation of the policy candidate,
wherein the second score is determined based on a time-derivative of an accumulation of lateral error between the forward simulation and a reference trajectory associated with the policy candidate; and
multiplying the first score and the second score to produce an aggregate score; and
controlling driving of the autonomous vehicle based on the respective set of scores of each policy candidate of the set of multiple policy candidates for each agent.
10. The method of claim 9 , further comprising:
determining a classification of each agent of the set of agents, wherein the set of multiple policy candidates is determined based on the classification.
11. The method of claim 9 , wherein operating the autonomous vehicle based on the respective set of scores of each policy candidate comprises: operating the autonomous vehicle based on a plurality of ego vehicle simulations.
12. The method of claim 11 , wherein each ego vehicle simulation of the plurality is based on an environmental scenario comprising a policy selected for each agent of the set of agents in the environment, the policy for each agent selected from the set of multiple policy candidates for the respective agent based on the respective set of scores of each policy candidate.
13. The method of claim 11 , wherein the forward simulation of each policy candidate is independent, wherein each ego vehicle simulation of the plurality comprises a combined simulation of the ego vehicle and each agent in the environment.
14. The method of claim 9 , wherein the respective set of scores of each policy candidate collectively characterizes a respective policy likelihood across both a retrospective observation period and a prospective prediction period.
15. The method of claim 14 , wherein each first score characterizes a historical comparison over the retrospective observation period, wherein each second score characterizes a predicted feasibility over the prospective prediction period.
16. The method of claim 15 , wherein the prospective prediction period is associated with a predetermined traversal distance or a predetermined time period.
17. The method of claim 9 , further comprising: using a multi-policy decision-making (MPDM) system, determining an ego policy for the autonomous vehicle based on the respective set of scores of each policy candidate of the set of multiple policy candidates for each agent, wherein operating the autonomous vehicle comprises executing the ego policy.
18. The method of claim 1 , wherein the lateral error is determined with respect to a lane centerline.