Optimization-Based Whole-Body Control of a Series Elastic Humanoid Robot

As whole-body control approaches begin to enter the mainstream of humanoid robotics research, there is a real need to address the challenges and pitfalls encountered in hardware implementations. This paper presents an optimization-based whole-body control framework enabling compliant locomotion on THOR, a 34 degree of freedom humanoid featuring force-controllable series elastic actuators (SEAs). Given desired momentum rates of change, end-effector accelerations, and joint accelerations from a high-level locomotion controller, joint torque setpoints are computed using an effcient quadratic program (QP) formulation designed to solve the floating-base inverse dynamics (ID). Constraints on the centroidal dynamics, frictional contact forces, and joint position/torque limits ensure admissibility of the optimized joint setpoints. The control approach is supported by an electromechanical design that relies on custom linear SEAs and embedded joint controllers to accurately regulate the internal and external forces computed by the whole-body QP. Push recovery and walking tests conducted using the THOR humanoid validate the effectiveness of the proposed approach. In each case, balancing is achieved using a planning and control approach based on the time-varying divergent component of motion (DCM) implemented for the first time on hardware. We discuss practical considerations that led to the successful implementation of low-impedance whole-body control on our hardware system including the design of the robot's high-level standing and stepping behaviors and low-level joint-space controllers. The paper concludes with an application of the presented approach for a humanoid firefighting demonstration onboard a decommissioned US Navy ship.