Human-Aware Control for Physically Interacting Robots.

This paper presents a novel model for predicting human movements and introduces a new control method for human-robot interaction based on this model. The developed predictive model of human movement is a holistic model that is based on well-supported neuroscientific and biomechanical theories of human motor control; it includes multiple levels of the human sensorimotor system hierarchy, including high-level decision-making based on internal models, muscle synergies, and physiological muscle mechanics. Therefore, this holistic model can predict arm kinematics and neuromuscular activities in a computationally efficient way. The computational efficiency of the model also makes it suitable for repetitive predictive simulations within a robot's control algorithm to predict the user's behavior in human-robot interactions. Therefore, based on this model and the nonlinear model predictive control framework, a human-aware control algorithm is implemented, which internally runs simulations to predict the user's interactive movement patterns in the future. Consequently, it can optimize the robot's motor torques to minimize an index, such as the user's neuromuscular effort. Simulation results of the holistic model and its utilization in the human-aware control of a two-link robot arm are presented. The holistic model is shown to replicate salient features of human movements. The human-aware controller's ability to predict and minimize the user's neuromuscular effort in a collaborative task is also demonstrated in simulations.