Model Predictive Control with Environment Adaptation for Legged Locomotion

Project Overview

Re-planning in legged locomotion is vital for tracking desired user velocities, adapting to terrain irregularities, and rejecting external disturbances. This project introduces and validates a real-time NMPC scheme that allows legged robots to traverse diverse environments.

Key contributions include:

• Mobility-Based Cost Function: Maximizes leg mobility to enhance adaptation to terrain features.
• Real-Time Re-planning: Utilizes a real-time iteration scheme for online re-planning at 25 Hz with a 2-second horizon.
• Efficient Dynamics: Employed a single rigid body dynamic model to boost computational efficiency.

Validated through simulations and real-world experiments on the HyQ robot, demonstrating capabilities in omni-directional walking and rough terrain traversal.