Biologically Inspired Deadbeat Control for Running: From Human Analysis to Humanoid Control and Back

摘要

This paper works toward bridging the gap between observations and analysis of human-running motions, i.e., motion science and robust humanoid robot control. It is based on the concept of biologically inspired deadbeat (BID) control, which facilitates both 3D running on flat ground and on 3D stepping stones. Further contributions include explicit foot step targeting during running, leg crossover avoidance, and the embedding of BID control into a quadratic-program-based whole-body controller. The controller is based on the encoding of leg forces and center-of-mass (CoM) trajectories during stance as polynomial splines, allowing for intuitive and purely analytical controller design. It allows a real-time implementation, is highly robust against perturbations, and enables versatile running patterns. This paper provides a method for purely analytical foot-step targeting, introduces a new method to increase kinematic feasibility on complex robot models, and presents advanced whole-body running simulations, including high-speed running and push recovery. The paper closes the circle to human motion science by comparing BID-based CoM trajectories and ground reaction forces to data from human-running experiments.