VagueBench: Benchmarking the Robustness of Embodied Agents on Ambient Context Illusions
This work identifies a novel phenomenon termed Ambient Context Illusion, where agents fail to execute tasks correctly when non-essential contextual details are omitted, despite unchanged task semantics, and demonstrates that this phenomenon persists in real-world robotic systems, where it can lead to incorrect decision...
Yang Qu, Qiang Fu, Zong-Hao Ying et al.
· 0 citations