Tanmay Ambadkar

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.

Read the paper ยท View code