When Should a Robot Think? Resource-Aware Reasoning via Reinforcement Learning for Embodied Robotic Decision-Making
The episode discusses the paper "When Should a Robot Think?" which introduces RARRL, a framework allowing LLM-based agents to manage their own cognitive resources. Agents learn when to execute actions versus when to reason, optimizing computational effort based on cost and benefit. This adaptive method significantly reduces inference time and improves task success rates in robotics.