Semantic Action RL

Semantic Action RL (SARL) is a UC Berkeley research software/algorithmic tool for adapting generalist robot policies by running reinforcement learning over a vision-language-action model’s language prompts rather than low-level robot actions.