面向脑控仿人机器人多任务控制的脑-机联合决策研究

Brain-machine decision fusion (BMDF) is to integrate a brain-computer interface (BCI) and a robot system in the decision-making level. It is not only a helpful way to promote development of the rehabilitation service for the movement disabled, but also of important value in military fields. In the brain-controlled humanoid robot system for multitask-performing, there are two major challenges for the BMDF methods, i.e., how to combine the decisions of both the BCI and the robot to generate more commands for controlling the humanoid robot, and how to deal with decision conflicts from both parts to acquire a higher accuracy rate..This study aims to develop a stable and reliable BMDF system via two stages. The first is to increase the control commands by innovatively integrating environmental interaction and hybrid modal to make the joint decisions according to application characteristics of the brain controlled humanoid robot system. This helps clarify the key issue of brainwave feature’s representation in the cognitive processes of steady-state visual evoked potential (SSVEP) and motor imagery (MI) influenced by the BMDF paradigm. The second is to improve the accuracy rate by evaluate the uncertainty of BCI decisions based on subject’s physiological and psychological states. Then we propose a novel method to improve the reliability of robot’s trajectory planning by processing conflicts on decisions of the brain and the machine. This helps reveal the key issue of coupling relationship between the uncertainty of the BCI decision and the real-time state of the subject.

脑-机联合决策是将脑机接口与机器人在决策层面进行融合，其成果不仅有助于推进运动障碍病人康复事业的发展，而且具有重要的军事价值。但是对于脑控仿人机器人系统的多任务控制需求，1）如何结合两者的输出决策以扩展指令数量，以及2）如何处理两者决策的冲突以提高正确率，仍是制约其研究和应用的主要挑战。.本项目拟从两个方面开展脑-机联合决策研究：根据脑控仿人机器人的应用特点，创造性地融合了环境交互和混合模式进行脑-机联合决策，有望阐明脑-机联合决策范式影响下SSVEP和MI两种认知过程的脑电特征表征，从而扩展脑-机联合决策系统的指令数量；首次在被试生理、心理状态的约束下度量脑机接口决策的不确定性，创新地通过脑-机决策的冲突处理来提高脑控仿人机器人轨迹规划的可靠性，有望揭示脑机接口决策不确定性与被试状态约束之间的耦合关系，从而提高脑-机联合决策系统的正确率。

本项目主要围绕脑控仿人机器人系统中的脑机联合决策问题，对脑机联合决策范式下的脑电特征解码和脑状态约束下的联合决策方法开展了深入研究。本项目的主要成果包括：考虑脑控机器人对智能家居设备的操作需求，设计了融合增强现实技术的混合模式脑机接口范式，分析了范式参数对混合模式脑机接口分类性能的影响；针对所用脑电信号模式的低信噪比问题，从特征-决策两个层面发展了混合模式脑电信号的高性能解码方法，提出了基于决策选择策略的稳态视觉诱发电位解码方法、事件相关电位脑电信号的高频特征解码方法、基于稀疏共空间模式滤波和正则化线性判别分析的运动想象电位解码方法；考虑到被试状态对脑机接口决策不确定性的显著性影响，分析了专注状态下脑电信号的个性化频带和复杂网络拓扑特征，提出了脑机接口操作过程中的专注脑状态实时估计方法；提出了融合注意力检测和频率识别的脑机联合决策方法，构建了专注状态约束下的脑机联合决策系统。本项目的开展为相关研究提供了高性能的脑机接口解码算法和专注状态估计算法，有助于解决范式影响下的脑电特征表征、专注状态约束下的联合决策等关键问题，相关研究成果为研究脑机联合决策提供新的思路和算法，对于推动脑控机器人研究的实用化进程具有重要的意义。