The project aims also to develop new methods for modelling complex whole-body movement in terms of extracted motion primitives extracted from kinematic data collected by motion capturing human motions. We are currently extending a method for the online synthesis of complex body movements based on the combination of learned dynamic primitives → in our laboratory (e.g. Giese et al. 2009; Mukovskiy et al. 2013), based on online reactive synthesis of whole-body movements, towards the real-time  kinematic control of real robotics platforms (bipedal humanoids).  The new architecture incorporates the sequential generation of multi-step sequences relying on the feedback-controlled switching of the spatio-temporal kinematic synergies.

Caption: comparison of AUTOMATICALLY generated walking-reaching-opening-grasping movement sequence, (left) retargeted to the bipedal humanoid HRP2, (right) generated using avatar with the original human subject skeleton.