CLOVER: Closed-Loop Value Estimation & Ranking for End-to-End Autonomous Driving Planning

Researchers introduced CLOVER, a closed-loop generator-scorer framework for autonomous driving planning.

CLOVER trains on evaluator-filtered pseudo-expert trajectories and predicts planning sub-scores to rank candidate paths.

It then uses conservative closed-loop self-distillation to refine proposals and better align training with evaluation.

The method reports new best or matching results on NAVSIM and NavHard, plus strong open-loop performance on nuScenes.