面向多任务人机协作的通道匹配机理研究

As intelligent technology, especially in artificial intelligence technology, develops rapidly, it is more and more urgent and important to satisfy demands of hybrid intelligent systems via the efficient human-computer collaboration, which appears in various fields, such as intelligent manufacturing, military, aeronautics and astronautics, etc. Multichannel human-computer interaction technology, as a core technology element of human-machine cooperation, has problems on both how to describe human behavior semantics (HBS) and how to match channels based on HBS in the process of multitask execution, which is on the basis of psychology and behavioral science. Based on the related theories, methods and techniques in information science, management science and behavioral science, the characteristics of human’s channel behavior are integrated into multichannel interactive scenes, and the behavior semantics of the multichannel signal features are both extracted and characterized. Furthermore, the association mechanism between behavioral semantics and tasks are researched. On this basis, we design a scheduling model of muti-objective optimization and its algorithm in matching between tasks and channels. Meanwhile, the mechanism of channel conflict decision is proposed and the prototype system is further designed to validate the theories and methods for channel matching among multitasks human-machine cooperation. The study of this project aims to provide a scientific theoretical basis for solving the channel matching problems in multitasks, and contributes to the great efficiency, nature and timeliness of human-machine cooperation as well as the mutual trust between human and machine.

随着智能技术特别是人工智能技术的飞速发展，通过高效人机协作以满足混合智能系统的需求，在智能制造、军事、航空航天等领域显得愈发迫切和重要。作为人机协作的核心技术要素之一，多通道人机交互在面向多任务执行过程中，存在如何结合心理学和行为科学刻画人的通道行为语义及基于行为语义的通道匹配问题。本项目在信息科学、管理学和行为科学等学科相关理论、方法与技术的基础上，将人的通道行为特征融入到多通道交互环境中，通过行为实验、信号处理和数据分析等研究多通道信号特征的行为语义提取及表征，探索行为语义与任务的关联机理及其相互关系；在此基础上，设计任务与通道匹配的多目标优化调度模型、求解算法及通道冲突决策机制；进而开发原型系统，对面向多任务人机协作的通道匹配理论与方法进行验证。通过本项目研究期望能为实现面向多任务的通道匹配问题解决提供科学的理论依据，为提高人机协作的高效性、自然性和时效性及人与机的互信任做出贡献。

本项目面向信息物理系统（CPS）中多任务人机协作场景，针对复杂决策问题求解理论和技术的智能化需求，在信息科学、管理学和行为科学等学科相关理论、方法与技术的基础上，系统研究了多任务人机协作中多通道匹配的关键理论、方法和技术。主要研究成果包括：.（1）构建了眼动、手势、语音的多通道行为库，提出了一种多通道行为语义的表征技术，面向多无人平台操控任务实现了多通道信号特征的提取与语义表征；.（2）面向人机协作复杂决策任务需求，提出了一种变粒度的任务分解方法，依据子任务粒度、子任务关联度和资源均衡度实现不同场景的任务可变粒度分解，为多任务实施奠定了方法基础；.（3）考虑行为者的主观风险偏好，构建了一系列三支行为决策模型，为多通道匹配奠定了理论基础；设计了面向多任务的通道能力评估指标体系，并以通道能力最大化且通道占用数尽可能少为目标，构建了任务与通道匹配的多目标优化调度模型及通道冲突消解机制； .（4）设计了一种多通道的智能切换方法，实现了交互过程的通道智能切换；进而分别设计并实现了“一人单机”、“一人双机”、“一人多机”的多通道交互系统，实现了“一人多任务”时片分布操控技术；.（5）提出了一种多通道交互分层处理通用模型，为多任务多通道人机交互系统的开发与设计提供通用的模型方案；进而将多通道交互应用于智能博弈系统，并基于注意力机制实现了智能体的关注热点以及行为语义生成。.本项目的成果已在月面探测航天员—机器人协调感知与交互、战术级信息智能协同交互、多无人平台协同任务规划等场景中得到初步的转化应用。为实现面向多任务的通道匹配问题解决提供科学的理论依据，为提高人机协作的高效性、自然性和时效性及人与机的互信任做出贡献，在融合人与机器智能的指挥与控制决策领域具有广泛的应用前景。