GAVEL Uses Graph World Models to Verify Long-Horizon Robot Planning
September 18, 2026
GAVEL introduces a graph-based world model to represent object relations and action preconditions for LLM-driven robot planning. The framework predicts action consequences to detect violations and repairs errors using the graph, reserving LLM replanning only for complex semantic reasoning.
HOW THIS AFFECTS YOU
●
builderYou can reduce LLM replanning costs and increase plan reliability by using a graph model for error detection.
●
researcherThis approach provides a structured way to combine probabilistic belief states with LLM planning.