BiGym 2.0: Benchmarking Learned and Agent-Developed Policies for Humanoid Household Manipulation
Paper • 2610.07594 • Published • 1
We develop safe and intelligent robot learning systems that leverage their entire physical embodiment. Our work spans online RL, offline RL, and imitation learning, with a focus on sample efficiency and safety — toward robots that learn efficiently, act intelligently, and operate safely alongside humans.