AutoSpec: Specification Refinement for Reinforcement Learning
AutoSpec turns coarse or under-specified logical objectives into refined, guidance-rich specifications that help reinforcement learning agents solve complex tasks.
Research summary
The framework monitors learning, diagnoses failure modes, refines specifications, and retrains agents to make difficult long-horizon tasks more learnable.