1 paper · 1 filter
Qinghang Liu, Conglin Wang, Zongsheng Li +6
Long-horizon robot manipulation requires a robot to remember task-relevant information beyond its short observation window. Existing memory-augmented policies typically retrieve re…