Adaptive Force-Based Control of Dynamic Legged Locomotion over Uneven Terrain

TL;DR

This work tackles robust dynamic locomotion for legged robots under significant model and terrain uncertainties, especially when carrying unknown loads. It introduces an adaptive force-based control framework by integrating $L_1$ adaptive control with both a force-based balance controller and a Model Predictive Control (MPC) backbone to compensate persistent disturbances and unknown terrain impact models. The approach enables heavy-load operation (up to $50\%$ of body weight) across rough and sloped terrains and supports multiple dynamic gaits (e.g., trotting and bounding) through a real-time dual-MPC computation scheme with a fast adaptive MPC at $300$ Hz and a slower reference MPC at $30$ Hz. Hardware validation on a Unitree $A1$ and extensive simulations demonstrate improved trajectory tracking and stability compared to non-adaptive baselines, highlighting practical impact for rescue, inspection, and logistics in unstructured environments. The framework’s ability to handle unknown terrain models without re-tuning and its extension to various gaits mark a significant advance in robust, real-time legged locomotion.

Abstract

Agile-legged robots have proven to be highly effective in navigating and performing tasks in complex and challenging environments, including disaster zones and industrial settings. However, these applications normally require the capability of carrying heavy loads while maintaining dynamic motion. Therefore, this paper presents a novel methodology for incorporating adaptive control into a force-based control system. Recent advancements in the control of quadruped robots show that force control can effectively realize dynamic locomotion over rough terrain. By integrating adaptive control into the force-based controller, our proposed approach can maintain the advantages of the baseline framework while adapting to significant model uncertainties and unknown terrain impact models. Experimental validation was successfully conducted on the Unitree A1 robot. With our approach, the robot can carry heavy loads (up to 50% of its weight) while performing dynamic gaits such as fast trotting and bounding across uneven terrains.

Figures (9)

• Figure 1: Our proposed adaptive MPC is successfully validated in an experiment on a Unitree A1 robot while carrying an unknown load of 5 kg (almost 50% of body weight) on rough terrain. Experimental results video: https://youtu.be/5t1mSh0q3lk.
• Figure 2: Baseline Control Structure. Block diagram of a control architecture for a quadruped robot. For the stance leg control, we use two common baseline control systems: QP-based balancing controller and MPC.
• Figure 3: Proposed adaptive force-based control system diagram. a) The main structure of the proposed adaptive force-based control system, b) Block diagram of the proposed adaptive QP-based balancing controller, c) Block diagram of the proposed adaptive MPC. Each dashed line indicates the update frequency for control components
• Figure 4: Motion snapshot of the robot with bounding gaits. A simple controller cannot easily predict The quadruped's center of mass motion (yellow line). This illustration can represent the importance of using MPC for reference model
• Figure 5: Navigating different terrain using our proposed adaptive MPC while carrying an unknown heavy load. a) gravel, b) grass, c) rough terrain, d) high-sloped terrain.