面向人机序贯决策实现共享控制下的仲裁优化

Shared control with optimized arbitration for human-machine sequential decision-making

共享控制存在于众多由人类智能和机器智能共同参与的序贯决策场景.由于人的决策范围和智能机器的决策范围尚未予以明确划分,需要加以实时仲裁从而达到人机共存并且共享决策权限.为此本文提出了一种仲裁优化方法,该方法的独特之处在于引入自主性边界概念,优化了共享控制中人机决策动作的仲裁机制.本文为自主性边界的计算和更新维护提供了思路,能够基于贝叶斯规则的意图推理分析人机共享系统可能要实现的目标,从而确定仲裁参数.此外,本文还分析了自主性边界的不确定性以促进边界信息对共享控制中决策质量的优化效果.实验结果表明,所提出的方法在累积奖励、成功率、撞击率方面表现出色,这些说明了本文提出的共享控制中的仲裁优化方法在求解人机序贯决策问题时的有效性和价值.

Human-machine shared control occurs in numerous decision-making scenarios involving both human and machine intelligence.Given the absence of a clear division in decision-making domains between humans and intelligent machines,there arises a need for real-time arbitration.This is essential to facilitate the coexistence of humans and machines and the equitable sharing of decision-making authority.Thus,this study introduces a unique arbitration optimization method.It distinguishes itself by incorporating the concept of an autonomous boundary into the optimization of the arbitration mechanism governing human-machine decision-making within shared control scenarios.This study presents a framework for calculating,updating,and maintaining this autonomous boundary.It leverages Bayesian rule-based intention inference to analyze potential goals within human-machine shared systems.This analysis aids in determining the selection of arbitration parameters.Furthermore,this study analyzes the uncertainty associated with the autonomous boundary,which aims to enhance the optimal impact of boundary information on decision quality within shared control.The experimental results demonstrate the effectiveness and value of our proposed method for solving sequential decision-making problems.This is evidenced by strong performance across various metrics,including cumulative reward,success rate,and crash rate.These results underscore the value and utility of our arbitration optimization approach for shared control in addressing sequential decision-making problems.