New MR-NMPC lets Unitree A1 walk on walls with 2.9x better success
Quadruped robot walks on two legs using walls for support 2.9x more reliably.
Deep Dive
A multi-rate nonlinear model predictive control (MR-NMPC) framework enables quadrupedal robots like the Unitree A1 to perform wall-supported bipedal locomotion. The system simultaneously plans contact points and center-of
Key Points
- MR-NMPC simultaneously optimizes contact points and CoM trajectories for wall-assisted bipedal locomotion
- Achieves 2.9x higher success rate on rough terrain vs. heuristic-based MPC
- Validated on Unitree A1 robot with robust performance under external disturbances
Why It Matters
Enables quadruped robots to navigate tight, unstable environments by walking on two legs with wall support.