The most frustrating part of imitation learning is collecting huge amounts of teleop data. But why teleop robots when robots can learn by watching us?
Introducing Point Policy, a novel framework that enables robots to learn from human videos without any teleop, sim2real, or RL.
