Researchers have proposed a framework called Evidence Acquisition and Feasibility Gating (EAFG) to improve the performance of robots executing long-horizon manipulation tasks from natural-language instructions. EAFG combines Vision-Language Models with Task and Motion Planning, allowing robots to acquire visual evidence through exploratory subgoals and decide whether to proceed with task planning or halt based on feasibility. Experiments show that EAFG improves recipe completion in cooking tasks by discovering task-relevant objects before planning, and reduces repeated attempts when an object is absent. --- Why it matters: This matters because it addresses a common challenge in robotics: partial observability, where the availability of goal-relevant objects may be uncertain. EAFG's ability to acquire visual evidence and make informed decisions can improve the reliability and efficiency of robots executing complex tasks from natural-language instructions. Source: https://arxiv.org/abs/2608.20084