Enchanced Torque Controlled Balancing through Multimodal Sensor Fusion based on an Extended Kalman Filter

Towards achieving dynamic humanoid robots that can function effectively in our unstructured environments, incorporation of some form of mechanical compliance and whole-body torque control present a very promising approach. However, one of the primary limitations affecting this approach is that of accurate, and effective state information, in particular, that of estimation of external forces that influence the body. Despite the presence of a rich suite of sensors in modern bipedal humanoid robots such as the iCub, a computationally efficient framework for real-time state estimation is still and open challenge. In this paper, we tackle a crucial subproblem of the walking and balancing challenge, that of accurate estimation of foot state and contact wrenches, in order to suitably facilitate torque-controlled balancing on compliant (soft) surfaces. Our approach, based on an Extended Kalman Filter (EKF), estimates the foot state as well as the wrenches it is subject to, in real-time. This estimation framework is then combined with the dynamic control task of torque-controlled balancing on a compliant surface introduced by a distributed tactile sensor (skin). We believe that this approach is an appropriate strategy towards enhancing the capability of torquecontrolled humanoids in coping with uneven and compliant terrains.