We have all seen humanoid robots that can recover from a while. Teaching them to stand up in can be challenging. HoST is a reinforcement learning framework that learns standing-up control from scratch. As the researchers explain:
HoST effectively learns posture-adaptive motions by leveraging a multi-critic architecture and curriculum-based training on diverse simulated terrains.
For testing this approach, a Unitree G1 robot was used. The robot was able to get up on a grass slope, when leaning against a tree, or other surfaces.
[HT]