Autonomous Vehicle Motion Planning for Risk-Aware Agent Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion planning methods for autonomous vehicles (AVs) are inefficient in handling high-risk but low-probability agent behaviors, requiring extensive computational resources to ensure robustness, and may lead to suboptimal decisions due to insufficient sampling of rare but dangerous scenarios.
Innovation Solution
A computer-implemented method that biases the selection of ego actions towards higher rewards while sampling agent behaviors from a risk-aware distribution, prioritizing riskier actions and behaviors to reduce the number of simulations needed, using techniques like importance sampling and multi-armed bandits to enhance robustness with reduced computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If extensive simulations are performed to ensure robustness against high-risk agent behaviors, then decision-making robustness is improved, but computational resources are excessively consumed
Solution Approach 1:
The patent changes the sampling distribution parameter from uniform to risk-aware, where the sampling probability is adjusted based on the risk level of each agent behavior. This allows the system to focus computational resources on high-risk scenarios while maintaining robustness, resolving the contradiction between reliability and computational resource consumption
Solution Approach 2:
The patent applies different sampling strategies to different agent behaviors based on their risk characteristics. High-risk behaviors are sampled more frequently while low-risk behaviors are sampled less frequently, creating a non-uniform sampling distribution that optimizes computational resource allocation across different behavior types
2Adaptability or versatility
If uniform sampling of agent behaviors is used, then all behaviors are explored equally, but high-risk low-probability behaviors are insufficiently sampled
Solution Approach 1:
The patent modifies the sampling probability parameter from a uniform distribution to a risk-aware distribution. The sampling probability for each agent behavior is changed based on its risk level, ensuring that high-risk behaviors are sampled more frequently while maintaining coverage of the full behavior space
Solution Approach 2:
The patent makes the sampling distribution dynamic by adjusting sampling probabilities based on risk assessments of different agent behaviors. This dynamic approach allows the system to adaptively focus on high-risk scenarios while maintaining comprehensive behavior coverage, resolving the contradiction between versatility and reliability
3Reliability
If the number of search steps is increased to improve sampling adequacy, then robustness to high-risk behaviors is improved, but planning time increases
Solution Approach 1:
The patent changes the sampling distribution parameter to risk-aware, which improves sampling adequacy for high-risk behaviors within a fixed number of search steps. This parameter change allows the system to achieve better robustness without increasing planning time, as the improved sampling efficiency compensates for the fixed computational budget
Data Source
AI summary
A computer-implemented method of planning ego actions for a mobile robot in the presence of at least one agent, comprising: searching for an optimal ego action in multiple search steps, each comprising: selecting an ego action from a set of possible ego actions, selecting an agent behaviour from a set of possible agent behaviours, running a simulation based on the selected ego action and agent behaviour, determining a possible outcome, and assigning a reward to the selected ego action, based on a reward metric, wherein selection of the ego action in later search steps is biased towards higher reward ego action(s) but selection of the agent behaviour in later search steps is biased towards riskier agent behaviour(s), a risky agent behaviour being, according to earlier search steps, more likely to result in a lower reward outcome and choosing an ego action based on the rewards computed in the search steps.