Search papers, labs, and topics across Lattice.
University of North Carolina at Chapel Hill
2
0
4
Robots struggle to match human-level reasoning in manipulation tasks, achieving only 16.3% success in a new benchmark that evaluates behavior-grounded reasoning.
Even with a near-optimal offline Q-function, adapting to a new environment can be as hard as learning from scratch, unless your offline data satisfies a new structural condition that this paper identifies.
