Augmenting Robot Teleoperation with Shared Autonomy via Model Predictive Control

Abstract

Shared autonomy enabled teleoperation systems minimise the cognitive load on an operator by providing autonomous assistance during task execution. In contrast to prior approaches using policy blending methods that employ a predict-then-act principle where the robot takes over when confidence in a goal is high, our proposed approach involves continuous policy adaptation. This approach utilises the augmented state of the robot, incorporating both the operator’s inputs as well as the robot’s autonomous assistance, to provide final assistive control to the robot. To address the issue of the operator’s trust in the robot, we formulate the approach as an optimal control problem with the objective of following the operator’s input commands while simultaneously adapting the user’s inputs to complete the task. We employ a Model Predictive Control (MPC) framework to solve this problem. We evaluated this framework through a user study on multiple goal picking tasks and compared it against pure teleoperation and proximity-based assistance methods. The results of the study show superior performance of our approach over the other methods in terms of trial completion times, collision avoidance, perceived ease of use, and responsive behaviour, indicating its effectiveness in improving teleoperation performance while maintaining user trust in the system.