Learning-Deep-Learning

MARC: Multipolicy and Risk-aware Contingency Planning for Autonomous Driving

June 2024

tl;dr: Generating safe and non-conservative behaviors in dense dynamic environment, by combining multipolicy decision making and contigency planning.

Overall impression

This is a continuation of work in MPDM and EUDM. It introduces dynamic branching based on scene-level divergence, and risk-aware contingency planning based on user-defined risk tolerance.

POMDP provides a theoretically sounds framework to handle dynamic interaction, but it suffers from curse of dimensionality and making it infeasible to solve in realtime.

MPDM and EUDM are mainly BP models, but MARC combines BP and MP.

belief trees heavily and decomposes POMDP into a limited number of closed-loop policy evaluations.

For the policy tree (or policy-conditioned scenario tree) building, we can see how the tree got built with more and more careful pruning process with improvements from different works.

All previous MPDM-like methods consider the optimal policy and single trajectory generation over all scenarios, resulting in lack of gurantee of policy consistency and loss of multimodality info.

Key ideas

Technical details

Notes