RISE lets humanoid robots steer how they get back up
Can a humanoid robot choose how it gets back up?
A humanoid robot that has fallen can now be steered toward how it gets back up, not just whether it does.
The same fallen pose can admit several ways to stand, with different routes, support transitions, and final headings. Getting up reliably is one problem. Deciding which way to get up is another, and a new robotics paper goes after the second.
The method is called RISE. It lets task objectives steer the recovery at inference time. A goal can be a heading, a rise speed, or a path.
It works in three steps. First, the robot learns a set of basic movements and how to carry them out using its sensory feedback. Different recoveries can be composed from those shared movements. Second, a visual diffusion planner learns from executions of those movements and predicts motion intent as future body states. Third, task costs guide that prediction toward a desired heading, rise speed, or path.
The authors report simulation experiments demonstrating diverse recoveries, terrain adaptation, and test-time steering without retraining. They also show real-world demonstrations on a Unitree G1.
The paper is by Osher Azulay, Cheng-Lin Hsieh and Stella X. Yu.
Which goal would you want a robot to pick when it gets up?
Archive video: Quantic Dream, CC BY 3.0 (Wikimedia Commons); Ferrante E, Turgut A, Duéñez-Guzmán E, Dorigo M, Wenseleers , CC BY 4.0 (Wikimedia Commons); NASA
Video: osheraz.github.io/RISE