Search papers, labs, and topics across Lattice.
1
0
3
Counterfactual human preference labels dramatically improve robot navigation safety, reducing collisions by nearly 50% in simulation and boosting real-world goal success by 24%.