Aggressive standing-up
Accepted at Humanoids. Optimized get-up maneuvers for humanoids on diverse terrain.
Robust and Optimized Aggressive Standing-Up Maneuvers for Humanoid Robots across Diverse Terrains. Accepted at the IEEE-RAS International Conference on Humanoid Robots (Humanoids).
Chuong Nguyen and Quang Huy Dao contributed equally, with Loc Pham, Truong Nguyen, and Quan Nguyen.
A full-size humanoid that falls on anything other than a lab floor still has to stand up. This paper optimizes contact timing and a whole-body trajectory for aggressive get-up maneuvers, including martial-arts-style kip-ups, then tracks that plan with a learned controller.
The video walks through the pipeline. Stage one takes an initial pose, a final pose, and a contact schedule, and optimizes timing and the trajectory. A real-time loop interpolates the reference and tracks it with joint-space PD control at 500 Hz. We compare that learned tracker with trajectory optimization paired with linear model predictive control and with nonlinear model predictive control, from both supine and prone starts. Hardware and simulation clips cover grass, scattered debris, and other uneven surfaces.