Humanoid robot learns to stand up under low obstacles
ROOF: Learning Humanoid Fall Recovery under Overhead Constraints
Robotics
Summary
Humanoid robots that fall under low ceilings or objects have trouble standing up because they need to make space without hitting the obstacle. The authors developed a learning method that guides the robot to gradually move out from under the obstacle and stabilize itself before standing. Their method uses only the robot’s own sense of position and joint status, and rewards reducing overlap with the obstacle without telling the robot exactly how to escape. This approach helped the robot recover successfully in most tested cases, including complex real-world setups like furniture.
What this means in practice
- •For robotics system integrators: Deploy robots in cluttered indoor environments where they can autonomously get up from falls under furniture or low obstacles.
- •For warehouse automation teams: Enable mobile robots operating under tight overhead spaces to recover after falls without external guidance or intervention.
Authors
Yidan Lu, Jiayuan Yu, Zongrui Cai, Jingwen Sun, Jinrui Han, Jing Wang, Ping Luo, Peng Lu
Abstract
A fallen humanoid beneath an overhead obstacle faces a coupled recovery problem. It must create clearance while organizing support for a stable rise, since attempting to stand before making room can further increase obstruction. We introduce a reinforcement learning method with contact-supported geometric escape guidance, which rewards incremental reduction of robot-obstacle overlap without prescribing an escape path. To achieve this, a learned motion prior, together with posture-dependent support and joint-stall costs, is proposed to shape a single policy that learns when to clear the obstacle, how to establish support, and when to complete the rise, all using proprioception alone. In simulation, it recovers in 88.5% of trials beneath plates that can only be lifted, compared with 18.9% without geometric guidance, and in 98.6% of trials beneath free plates with randomized size, mass and placement. 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. Project page: https://lu-yidan.github.io/ROOF/