Machine Learning Based Human-Robot Shared Control

In most of the current robotic systems, teleoperation dominates the mainstream and is the major form of control. However, most of the tasks contain a lot of relatively simple but repetitive sub-tasks, which brings substantial burdens to the operator. Therefore, it is necessary to bring autonomy to current robotic systems. For safety considerations, human-in-the-loop control is necessary. In this case, we aim to build a machine learning-based human-robot shared control framework, which can automate some of the sub-tasks via deep imitation learning or deep reinforcement learning.

Manual control mode is incorporated in the shared control framework to ensure that humans can take over the whole operation process at any time, while some of the sub-tasks can be performed autonomously. Thus, the strengths of the humans' judgments and the robot intelligence are combined to provide reliable and efficient control.