Active Dynamic Load Adaptation for Quadruped Locomotion on Complex Terrain
Yimin Xiao, Dianzhong Li, Wangjun Huang, Ying Sha, Li Qin
AI summary
Problem
Quadruped robots lose stability when carrying active dynamic loads that continuously shift mass and apply torques, especially when combined with terrain-induced disturbances. Existing methods cannot simultaneously adapt to these real-time load and terrain dual disturbances.
Approach
The method uses a reinforcement learning framework equipped with an inverse dynamics model to capture active load disturbances and a forward dynamics model to predict terrain effects, allowing real-time adaptation using only proprioceptive feedback.
Key results
- Reduced base roll deviation and improved velocity tracking under active load disturbances
- Successful sim-to-real deployment on a Unitree Go2 robot with a mounted mechanical arm
- Robust adaptation to varied load movements across flat, stair, rough, and slope terrains
- Outperforms baseline PPO and HIMLoco in stability metrics during dynamic load transitions
Why it matters
Enables safer and more capable load-carrying quadruped robots for logistics, search-and-rescue, and agricultural applications where terrain and payload dynamics constantly interact.
Abstract
Quadruped robots show important potential for load-carrying tasks due to their terrain adaptability, and a unique challenge of these tasks is to maintain quadrupedal stability when the load has active and dynamic characteristics. The load’s mass and center of mass change dynamically, rather than being integrated as a whole-body component of the quadruped. Unlike traditional load-carrying tasks, where the load is typically passive and its influence on the robot’s movement is predictable and static, active dynamic loads can actively alter the robot’s balance control in real-time, impos- ing load disturbances on the robot’s locomotion. These load disturbances, when combined with the fundamental attitude changes induced by complex terrain, introduce dual dynamic disturbances to the robot. To address these dual disturbances, we propose an active dynamic load modeling approach that captures the active and dynamic characteristics of the load, enabling the robot to adapt to the real-time changes in load movement. This approach is integrated into a Reinforcement Learning (RL) framework that leverages dynamic models: an Inverse Dynamic Model (IDM) which learns the dynamic characteristics of the active load, and a Forward Dynamic Model (FDM) which predicts the effects of complex terrain on the robot’s motion, enabling synchronous adaptation to both types of dynamic disturbances. Extensive comparative simulations and physical experiments across diverse terrains, with active dynamic load of varying movements, demonstrate the effectiveness of our method in enhancing balance control and adaptability. More details are available at: Project Page.