Paper Everyday 💡 Ever had your high-level instructions misunderstood by your robot planner? ISR-LLM uses a three-phase loop—translating natural language to PDDL, generating an initial plan with an LLM, then iteratively refining it with a validator before execution. The self-refinement leverages both LLM-based and external validators to spot infeasible actions and feed corrections back into the planner for continuous improvement. Tested on cooking, ball-moving, and blocksworld domains, ISR-LLM achieves success rates markedly higher than state-of-the-art LLM planners—all while preserving natural language flexibility. Let your robot think, critique, and perfect its own plans before moving a joint 🚀 arxiv.org/abs/2308.13724

Robots are the future.