Red-Blue Adversarial Game Framework for Robot Safety Policy Learning

A proposed framework pits a Red Team agent generating hazardous scenarios against a Blue Team agent refining safety policies, iteratively surfacing high-risk edge cases that random simulation misses. The paper is a problem formulation and architecture proposal, not yet an empirical evaluation.