Implementation of a Robust Dynamic Walking Controller on a Miniature Bipedal Robot with Proprioceptive Actuation

Developing a robust dynamic walking controller for bipedal robots remains challenging as the system is hybrid, highly nonlinear, and strongly restricted. The typical two-level structure of high-level footstep planning and low-level wholebody control has been proven an effective approach for bipedal locomotion. However, practical guidance on its implementation is rarely covered fully in detail. To bridge this gap, this paper presents a detailed implementation of such controller for dynamic walking applications on a miniature bipedal robot with proprioceptive actuation. To the best of our knowledge, this is the first fully-untethered miniature bipedal robot which can achieve robust dynamic walking using this framework. In particular, the high-level planner determines both the location and duration for the next few steps based on the divergent component of motion. The low-level controller leverages the fullbody dynamics to establish the foot contact as planned while regulating other task-space behaviors, e.g., center of mass height and torso orientation. Both problems are formulated as smallscale quadratic programs, which can be solved efficiently with guaranteed optimality for real-time execution. Extensive results of simulation and hardware walking experiments are provided to demonstrate the strong robustness of the approach under various disturbances and uncertainties, e.g., external pushes and irregular terrains.