Reference Trajectory Adaptation to Improve Human-Robot Interaction: A Database-Driven Approach

Many robotic devices that are used for therapy or as assistive devices rely on pre-defined reference trajectories to assist the user's movements. Fixed pre-defined trajectories force the user to adapt to unnatural movement patterns which may be detrimental to rehabilitation outcomes. We propose a database-driven approach to adapt the reference trajectory of robotic training devices that rely on cyclic motion such as walking. Dynamic time warping is used to compare the measured pattern with a database of pre-approved safe trajectories; the best matching pattern is selected from the database and used for the next movement sequence. The proposed approach was evaluated in computer simulations and a bioinspired robotic test bench. Our proposed method reduced the RMS error between individual user trajectories and the supplied reference, even in the presence of measurement noise.