Disney teaches a robot how to fall gracefully and make a soft landing

Bipedal (two-legged) robots are sophisticated machines, but they are not the most graceful when things go wrong. A simple push, fall or an obstacle can send them crashing to the ground, often resulting in expensive damage to sensitive components such as cameras.

To solve this problem, researchers at Disney Research in Zurich, Switzerland, have developed a new system that ensures that when gravity brings a robot tumbling down, it falls softly and gracefully.

Existing techniques to protect robots when they topple over do not offer control or effectively minimize impact. Actuators might freeze, causing the robot to stiffen and hit the ground hard, or they might go limp, leaving the robot to tumble chaotically. Other methods rely on pre-programmed falling motions, but these only work for slow movements or simple falls.

The AI solution

In a paper published on the arXiv preprint server, the researchers describe how they used reinforcement learning to train robots to fall softly and safely. They ran thousands of virtual robots inside a computer simulator and sent them tumbling from every conceivable angle and position.

For every attempted fall, the virtual robots received a reward for minimizing crash impact and for achieving a controlled landing in a target pose from a set of possible poses. The constant feedback ensured the robots learned a general strategy for executing a graceful crash, no matter how the fall began.

The result was a set of rules (or policy) that was transferred to a real bipedal robot for tests. The team selected 10 graceful end poses designed by artists, then pushed or tripped the robot with a stick to make it fall from random angles. After repeated falls, the robot showed no noticeable damage and remained fully functional. It also consistently executed controlled falling motions, finishing in the intended artist-designed poses.

“Our method enables soft falling behavior that protects the robot’s most sensitive part, regardless of the falling direction,” wrote the researchers in their paper. “This is the first general approach that demonstrates user-controlled falling of a bipedal robot in the real world.”

Next steps

The Disney team now wants to build on this breakthrough by addressing several challenges, including testing their AI strategy across different types of robots, such as four-legged machines, to determine whether it is universal.

They also want to develop a way for a robot to predict a fall, as the current system only takes over after a fall has started. Additionally, the researchers would like to design a way for a robot to get back up gracefully after hitting the ground.

© 2025 Science X Network