QP-based task-space hybrid / parallel control for multi-contact motion in a torque-controlled humanoid robot

Abstract

Humanoid robots rely on precise interaction force to locomote and perform various tasks. Controlling torque usually allows humanoid robots to produce these desired forces on known environments. However, the tracking may be imperfect in the absence of torque feedback or with an imprecise environment model. Furthermore, the presence of geometric errors, regarding the model of the environment, can also lead to discrepancies between desired and actual forces. In this paper, we extend our previous QP-based robust torque control framework to allow force control without requiring joint torque feedback. The control relies only on force/torque sensors at the end effectors, joint encoders and IMUs for kinematic feedback. Additionally, it is formulated to keep consistency with the internal state of the QP solver. We show that hybrid or parallel control, where position and force can be controlled independently, is possible with this approach. The framework is validated with stabilizer-free locomotion on uneven terrain and a multi-contact scenario with reference forces.