Computing robot internal/external wrenches by means of inertial, tactile and F/T sensors: Theory and implementation on the iCub

Reliable access to dynamics is one of the central challenges in humanoid robotics. In this paper we consider the problem of computing both internal and external wrenches in open multiple branches kinematic chains, which are not equipped with joint-level torque sensors, but on the contrary are provided with inertial and tactile sensing, and a set of six- axes Force/Torque Sensors (FTS) distributed within the chain. The proposed method is grounded on the Enhanced Oriented Graph, a graph-representation of the kinematic chain, enriched with the information coming from the different sensors. Under suitable conditions, a maximum of N+1 external wrenches can be estimated from N six-axes FTS. The graph is built dynamically, and internal and external wrenches are consequently updated. Theoretical results have been implemented in a software library (iDyn) released with an open-source license (GPL) as part of the iCub software project. The proposed method has been applied to 32 of the 53 DOF of the iCub humanoid robot.