Evidence-Gated Task and Motion Planning with Vision-Language Models
The Evidence Acquisition and Feasibility Gating (EAFG) framework helps robots perform complex tasks from natural language by first gathering visual evidence through exploratory subgoals guided by a Vision-Language Model (VLM). It then uses a feasibility gate to decide whether to proceed with planning, seek more evidence, or halt. This improves task completion by discovering necessary objects before planning and correctly handling situations where objects are missing.