Assistive Control of Robot Arms via Adaptive Shared Autonomy

Abstract

Shared autonomy is a robot control approach that assists human users to achieve their intended goal while leveraging the precision and efficiency of robot autonomy. In shared autonomy, user input and autonomous assistance are combined to effectively control robots without requiring users to provide direct and precise control inputs. A persistent question in shared autonomy is how to determine the arbitration between user input and autonomous algorithm. Due to variability in users’ desired amount of assistance, it is imperative to develop user-centric algorithms that provide customized and adaptive assistance by considering users’ preference, physical capability, and expertise. In this paper, we propose a shared autonomy method that factors in both users’ task performance and level of expertise to adaptively adjust the amount of assistance at runtime. We validated our method in an assistive control problem where human users teleoperate a robotic arm to perform object reaching and grasping tasks in a simulated environment. The results show that our method assisted the users to achieve a higher efficiency in accomplishing the object reaching and grasping tasks compared to direct teleoperation and two baseline arbitration methods that only consider task-related metrics.