FACTR 2: learning external force sensing for commodity robot arms
arXiv·medium signal
FACTR 2 (arXiv 2606.12406, Oh, Liu, Tao) tackles contact-rich manipulation on commodity robot arms that lack dedicated force sensors due to cost, learning to infer external forces and showing it improves downstream policy learning. By recovering force sensitivity without specialized hardware, it lowers the barrier for embodied-agent teams building manipulation policies on cheap arms. A practical contribution for the robotics-agent corner of the ecosystem where sensor budgets are tight.