Human Posture Prediction During Physical Human-Robot Interaction

When a human is interacting physically with robot to accomplish task, his/her posture inevitably influenced by the movement. Since not controllable, an active imposing collaborative trajectory should predict most likely posture. This prediction consider individual differences and preferences of movement execution, it necessary evaluate impact robot's action from ergonomics standpoint. Here, we propose method predict, in probabilistic terms, postures for given executed scenario. We formalize problem as joints velocity current end-effector velocity. The key idea our approach learn distribution null space Jacobian weights weighted pseudo-inverse demonstrated movements: both carry information about postural preferences, leverage redundancy ensure that predicted will be coherent position. validate simulated toy on two real human-robot interaction experiments where Franka robot.