Fall Recovery

A humanoid or legged robot's ability to stand back up on its own from lying or sprawled positions after falling.

Fall recovery means a legged robot, after falling, gets back up from lying on its back, lying face-down, lying on its side, or leaning against a wall, returning to a state where it can keep walking. Humanoid robots have a high center of mass and a small foot support area, so falling in the real world is hard to avoid entirely; if a person has to help it up every time, the robot cannot really be deployed autonomously. Getting up is harder to learn than walking: the torso and limbs touch the ground at multiple points at once, the order of contact is not fixed, and the reward signal is sparse. In the past, robots typically relied on hand-choreographed, fixed getting-up motions, which tend to fail when the posture or terrain changes. In 2025, methods emerged that train with reinforcement learning in simulation and transfer directly to a real Unitree G1 robot, such as HoST and HumanUP, both published at RSS 2025. A related concept is fall protection, minimizing impact and protecting the hardware during the fall itself.

ExampleTested on a Unitree G1, HumanUP lets the robot stand up on its own starting from either lying on its back or lying face-down, on grass, snow, and slopes.

Also called

Getting Up, Standing-up Control

Related

HoST (Humanoid Standing-up) · Fall Mitigation and Fall Recovery · Push Recovery · Balance Control · RL-based Locomotion Control · Humanoid Robot

Sources

Learning Humanoid Standing-up Control across Diverse Postures (HoST, arXiv 2502.08378)
Learning Getting-Up Policies for Real-World Humanoid Robots (HumanUP, arXiv 2502.12152)

As of

2025-04