Human-aware motion reshaping using dynamical systems

In this work, we present a real-time approach for human-aware motion replanning using a two-level hierarchical architecture. The lower level leverages stable dynamical systems to generate motor commands and to online reshape the robot trajectories. The reshaping strategy modifies the velocity of the robot to match three requirements: i) to increase the human safety in case of close interaction with the robot, ii) to guarantee the correct task execution in case of unforeseen obstacles (including the human), and iii) to replan online the current task taking into account the human behavior. The lower level has to execute all needed computations in real-time. To this end, we also propose a novel approach that leverages depth space structure and parallel programming to rapidly and accurately estimate the robot-obstacle distances. The higher level of the architecture monitors the human and provides to the lower level information about the human status. The proposed approach has been tested in a human robot interaction scenario, showing promising results in terms of safe human-robot coexistence.