Simulation of an Online Estimation Algorithm for Time-Dependent Kinematic Synergies: Towards Synergy Shaping

Synergies are a basis from which human motor behavior can integrate into robots for personalized feedback. However, uses of synergies in robotics remain limited to movement and pose reconstruction for robotic limbs or limb visualizations in virtual reality. In this paper, we present an algorithm to accurately, efficiently, and smoothly estimate the kinematic synergies of individual users. The algorithm is capable of online implementation due to iterative minimization, which uses a weighted cost function and constraints to determine the time-step and dilation of the users' synergies. To validate the method, we recruited three trained, able-bodied subjects to perform two bimanual VR tasks, peg transfer and knot tying, on a robotic exoskeleton. We recorded joint angle data and computed the synergies for each subject from training data. Then we used separate data from testing to simulate the online synergy estimation algorithm. The algorithm is accurate given a mean RMS movement estimation error of 6.8% for the current simulated time-step and 15.2% for the entire signal. The algorithm is also computationally efficient and provides a relatively smooth output, so it can be easily integrated into most modern systems. Therefore, in combination with robotic feedback, this online estimation algorithm has the potential to personalize the shaping of human motor behavior.