Addressing agent disagreement in mixed-initiative traded control for confined-space manufacturing

摘要

Control trading between fully automated robot control and teleoperated robot control can take advantage of both machine precision and speed and human adaptability and reasoning to improve overall performance. In mixed initiative interactions, both the human and machine can take control and execute the next action. However, when both the human and the machine are allowed to propose control trades, there is potential for conflict if there is no consensus on whether the trade should occur. The main contribution of this article is to integrate an explicit consensus development process that only completes a control trade with the consent of the agent that did not propose the trade and includes an explicit contingency procedure if consent is not given. This contingency-based mixed-initiative control trading is demonstrated with a robotic peg-in-hole insertion task that captures end-effector positioning challenges associated with confined-space manufacturing tasks such as drilling. Experimental results show that limit cycling (the rapid switching of control initiative), when the agents disagree on a control trade, can be removed by using a comprehensive set of task-specific contingency procedures. For the demonstration task, the proposed consensus-based approach facilitates the traded controller to i) substantially reduce baseline operation time by 50% when compared to the pure teleoperation case; and ii) increase success rate in the presence of sensing errors from 3% with fully automated operation to 97%.