Home▸Robotics & Kinematics▸MPC Legged Robot Balance

MPC Legged Robot Balance

Push a quadruped robot off balance and compare a naive reactive controller against a receding-horizon Model Predictive Controller: watch the MPC pre-plan a smooth multi-step footfall recovery from a predicted center-of-mass trajectory, while the reactive controller only corrects the current instant and can stumble.

Robotics & Kinematics3DAdvanced60 FPS

A quadruped robot walks forward while its sideways balance runs on a simplified inverted-pendulum model. Switch between a naive reactive controller — which only corrects the current-instant balance error — and a receding-horizon Model Predictive Controller, which re-solves a short future trajectory every step and pre-positions footfalls to absorb a disturbance smoothly. Fire a push, tune its strength, and watch the MPC's predicted center-of-mass ghost trail extend further ahead as you raise the horizon length, letting it shrug off pushes that topple the reactive controller.

⚙ Under the hood

Push a quadruped off balance and compare a naive reactive controller against a receding-horizon Model Predictive Controller: watch the MPC's predicted center-of-mass ghost trail extend further ahead as the horizon grows, pre-planning smooth multi-step footfalls that recover from disturbances the reactive controller topples over from.

Three.jsroboticsmodel predictive controlMPClegged robotbalance controlreceding horizon

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install