Robots are trained to get around obstacles and recover from falls. What happens when they get stuck under a table or other debris? ROOF explores that. It is a reinforcement learning framework that teaches G1 how to escape these situations. It has a reward system that guides the G1 as it overlaps less with obstacles and creates greater distances. Thanks to this approach, these robots can shift their body and use their limbs to get unstuck. As the researchers explain:
On a Unitree G1, the deployed policy recovered in 95 of 150 attempts across seven real-world obstacle scenes, including furniture and multi-object arrangements absent from training.
The robot can get up from under a long wooden box, ladder, and loose obstacles.
[HT]
{Authors: Yidan Lu and Jiayuan Yu and Zongrui Cai and Jingwen Sun and Jinrui Han and Jing Wang and Ping Luo and Peng Lu}
