Toward human-like motion generation in the Cartesian domain with robotic manipulators

The characterization of human upper limb kine- matics is fundamental not only in neuroscience and clinical practice, but also for the planning of human-like robot motions in collaborative robots. One promising approach to endow anthropomorphic robotic manipulators with human motion characteristics is to directly embed human upper limb principal motion modes at joint level, which are computed through functional analysis, in the robot trajectory optimization. This planning method poses some challenges when the kinematics of the manipulator is different from the model used for human data acquisition. In a previous work, we proposed to tackle this issue by mapping human trajectories onto robotic systems relying on Cartesian impedance control. An alternative method to overcome these problems could rely on the exploitation of the main human functional modes extracted directly in the Cartesian domain. Toward this direction, in this work we present the results of the functional characterization of upper limb kinematics in the Cartesian domain, considering data from 33 healthy subjects during the execution of daily- living activities. We found statistical differences between the amount of variability explained by a given number of basis elements in different directions of the Cartesian space. This suggests that some directions of the space are associated with a more complex motion evolution with respect to others, opening interesting perspectives for robot planning, neuroscience and human motion control.