AI robots solve puzzles better with blurry vision than perfect data

Original: Probing Embodied LLMs: When Higher Observation Fidelity Hurts Problem Solving

Source: arxiv.org ↗