Design of 3-D Operation Support System with Variable Autonomy via Gaussian Process Regression

摘要

In this paper, we design a human-robot collaboration system that supports 3-D manual reaching task of a robot manipulator to one of potential candidates of targets. This semi-autonomous robotic task is categorized into so-called overlapping interaction, where a human and an automatic controller determine the same signal and how to blend them is appropriately determined so that ideal operation supports for the operator are achieved. To this end, we build a human model from the operation data of an expert through Gaussian Process Regression (GPR), and design an autonomy determination mechanism based on the variance information given by GPR. Moreover, in order to allow the human interventions in dealing with various uncertainties in the real operation, we further add logic to switch automatic and manual control to the autonomy determination mechanism based on the variance of the operator's GPR model. Various user studies demonstrate the effectiveness of the present support system in terms of control performances and human workload.