Combining reduced dynamics models and whole-body control for agile humanoid locomotion

Abstract

In the near future, mobile machines are expected to leave the labs and perform more and more complex tasks in environments designed for human beings. Systems based on legged and especially bipedal locomotion have the potential of outperforming wheel-based locomotion systems in certain scenarios. Yet, at the current point in time, it still seems to be a long way to go until legged systems can unfold their full potential. Bipedal locomotion is a difficult controls problem due to issues such as underactuation, unilateral constraints and the hybrid nature of stepping. This work aims at first reducing the complexity of the considered problem as much as possible, while tackling the full problem in a second step. More explicitly: first, only the center of mass (CoM) dynamics of the robot is considered, which is a reduced but correct representation of its motion. Using this reduced model, analytical controllers are designed that include multi-step preview for both walking and running. The walking control method is based on the concept of Divergent Component of Motion (DCM), which extends the earlier concept of Capture Point to 3D. The proposed running control framework is called Biologically Inspired Dead-beat (BID) control. It is based on the encoding of CoM motions as polynomials during stance and explicitly solves for intuitively designed running boundary conditions. Both methods are powerful, purely analytical and very insightful. In a second step, the locomotion controllers are embedded into a quadratic programming (QP) based whole-body control framework. The latter allows for instantaneous optimal control, which ensures good trade-offs between the different necessary tasks at hand. Both control frameworks are tested extensively in simulation. Real experiments are successfully performed for DCM-based walking control and the whole-body control framework.