1 paper
Hengyan Liu, Wenlve Zhou, Bo Yue +7
Reactive vision-language-action (VLA) policies suffer from task-state aliasing in long-horizon manipulation, where identical multimodal inputs call for distinct, context-dependent…