Biomimetic and Psychophysical Investigations on Lifting Tasks for Developing Cooperative Reinforcement Learning Control of a Power Assist Robotic System

Different human-centric cooperative control strategies are used for enhancing human-friendliness in collaborative manipulation between a human user and a power assist robotic system (PARS). However, despite having tremendous prospects, investigations on developing machine learning-based cooperative controls for PARSs have not received much attention yet. It is hypothesized that a cooperative control strategy developed in the framework of reinforcement learning following biomimetic and psychophysical approaches may outperform the existing cooperative control methods for PARSs. As an initial effort towards developing a biomimetics and psychophysics-based cooperative reinforcement learning control strategy for PARSs, in this paper, we conducted a joint biomimetic and psychophysical study for object manipulation. To do so, we separately presented models of lifting objects manually and with a PARS considering weight perception, and kinematic and kinetic features (psychophysical approach), compared the model of manual lifting to that of power-assisted lifting (biomimetic approach), and conceptualized a cooperative reinforcement learning control framework for the PARS based on the biomimetic and psychophysical study results. The results showed that the perceived weights, kinetic features (peak load forces and load force rates), and kinematic features (peak velocities and peak accelerations) for manual lifting were higher than that for power-assisted lifting. A time delay between position and force trajectories was observed for power-assisted lifting, which was not observed for manual lifting. The findings were proposed to be used to develop a user-friendly cooperative reinforcement learning control framework for PARSs for handling large and heavy objects in various industries that could enhance human-robot interactions (HRI) and manipulation performance.