Motor learning of arm reaching movement in redundant musculoskeletal system

Abstract

Human can learn to move limbs in novel environments by modifying the endpoint trajectories, for example, correcting hand trajectories to accurately throw a ball to a target. However, the complicated modification at multiple joints is needed to correct the limb endpoint, since the endpoint movements are generated by the combination of multi-joint motion. Then, how is the learning response at a limb endpoint distributed among different joints? To answer the question, participants performed horizontal reaching movements to each of 8 targets while holding robotic handle by their right hand. For the training, the reaching movements were performed under the presence of a velocity-dependent curl force field. Learning response of hand force was quantified as the force pushing a randomly-interleaved virtual force channel, in which the movement trajectory of the handle was constrained to a straight path from the start position to the target. During the tasks, the torques in shoulder, elbow and wrist joints were estimated from the kinematic data obtained using a motion capture system. As a result, the hand movements were corrected by exponentially increasing the force against the force field across each training trial. During the training, muscle torques in each joint were gradually modified to generate the corrected force. Interestingly, the learning to the torque after the training trials was faster in the distal joint than in the proximal joints. Accordingly, early in the training the joint motion in the distal joint would be corrected to generate the desired hand movement whereas late in the training the proximal joint was modified to fine-tune the hand motion. Therefore, movements can be learned in novel environments by modifying the multi-joint motion in chains.