Biomechanics-Aware Dynamic Movement Primitives (Bio-DMP) for Adaptive Robotic Shoulder Physiotherapy

Abstract

Personalised rehabilitation is essential for restoring shoulder function following injury or surgery, due to anatomical complexity and variability across patients. While existing robotic systems offer consistent and intensive therapy, they often neglect internal musculoskeletal stress, such as tendon strain, which can potentially result in physiologically unsafe movements. The paper proposes Biomechanics-aware Dynamic Movement Primitives (Bio-DMP) for adaptive robotic shoulder physiotherapy. The therapy movements are learned through trajectories that operate in tendon strain space derived from musculoskeletal modelling using an OpenSim biomechanics model. Regions of elevated tendon strain on strain maps are identified and clustered as ellipsoidal high-strain zones. These zones exert repulsive virtual forces on the trajectory to avoid unsafe states, while an adaptive time scaling mechanism slows motion near unsafe areas to ensure smoothness. We validated the proposed method both in simulations with a manually drawn baseline trajectory and in real experiments with a trajectory recorded through kinaesthetic teaching using a KUKA LBR iiwa-7 robot. Both trajectories were adapted across six patient models with varying tendon insertion points. In all cases, the adapted trajectories avoided high-strain zones and remained smooth and dynamically feasible.