Cognitive Model Predictive Learning Cooperative Control to Optimize Electric Power Consumption and User-Friendliness in Human–Robot Co-manipulation

We developed a 1DOF PARS (power assist robotic system) for lifting heavy objects in collaboration with a human user. We considered human cognition (weight perception) when deriving the dynamics and control model for the system. A computational model for estimating electric power consumption in the system for the lifting task was derived. A cognitive model predictive control (MPC) was proposed that optimized electric power efficiency by optimizing the co-manipulation speed (i.e. by suggesting an optimum co-manipulation speed). The application of the proposed cognitive MPC showed a higher level of electric power efficiency. Human user’s psychological acceptance of the co-manipulation speed (i.e. user-friendliness) was learned applying a psychophysics-based reinforcement learning method, and then the MPC was redesigned to optimize the co-manipulation speed to result in optimum power consumption at optimum user-friendliness. The results we obtained can be used to develop predictive control strategies for human–robot collaborative tasks.