| Brain-computer interface(BCI)is a new type of human-computer interaction technology that establishes information pathways directly between the brain and the external environment without relying on peripheral nerve and muscle tissue.Human-computer collaborative hybrid augmented intelligence combines human cognitive and decision-making capabilities with the machine’s powerful computing and storage capabilities.It is seen as a new form of next-generation artificial intelligence.Human-computer shared control of the human-in-the-loop is an essential topic of hybrid intelligence research.As the highest form of interaction between brain intelligence and machine intelligence,BCI has unique advantages in coupling human analysis and response to uncertain problems into machine intelligence systems,so the research of BCI-based shared control technology has attracted extensive attention from scholars in recent years.With the goal of achieving efficient BCI-based human-machine shared control as traction,this dissertation explores the shared control methods in terms of system design,decision fusion methods,and BCI paradigms in the framework of time-sharing,local and global shared control using motor imagery(MI)-based BCI,steady-state visual evoked potential(SSVEP)-based BCI and high-resolution SSVEP-BCI,respectively,focusing on three dimensions: BCI target resolution,human-computer sharing degree and fault tolerance.The main research contents and results are as follows.Exploring time-sharing shared control based on target adaptation.With the typical application of BCI environmental control as the research background,a human-computer time-sharing shared control framework is designed to explore an efficient and intelligent environmental control method by embedding environmental understanding technology,asynchronous MI-BCI,and decision-making technology into the system.The designed environmental control system,oriented to rehabilitation hospital scenarios,integrates medical assistance calls,daily living assistance calls,and appliance control to enhance patients’ autonomy with motor function impairment.The MI detection method is improved,the user initiates control autonomously through the BCI module,the system combines the environment understanding module to infer human control intention adaptively,and the decision algorithm is used to dynamically update the interaction process by combining the environment state and human commands.10 subjects participated in online experiments in a simulated scenario,and the average response time for asynchronous detection was 3.38 s.The average accuracy of synchronous command recognition reached89.2%,demonstrating the practicality of the proposed environmental control system.Exploring local shared control based on opinion dynamics.The application of BCI technology in the multi-robot domain and the dynamic SSVEP-BCI based shared control approach are explored in the context of multi-robot foraging tasks.A human-robot shared control framework is proposed,which uses an opinion dynamics approach to combine human and computer commands into a consistent final decision,allowing multi-robot systems to get rid of the complete obedience to a single control source.For SSVEP-BCI,a dynamic stopping strategy is investigated to dynamically adjust the stimulus duration while ensuring the accuracy of SSVEP detection.In addition,a probability distribution form of human opinion expression is used to provide greater fault tolerance for human error commands.It has been experimentally verified that shared control with human-robot participation can perform better in large-area foraging tasks than multi-robot autonomous control or human control alone.Exploring global shared control based on high-resolution SSVEP-BCI.Large-scale area foraging tasks challenge the instruction set size of the BCI paradigm.To address this problem at the source,a high-resolution SSVEP-BCI paradigm with 112 targets is proposed for the shared control framework.Using the principle of competitive neuronal dynamics,40 stimulus frequencies are expanded to 112 targets through spatial multiplexing of visual stimulus frequencies.Three paradigms,aligned arrangement,nonadjacent arrangement,and row stagger arrangement,were designed with different stimulus arrangements.The global target localization method used the graph neural network algorithm,and six subjects participated in the offline validation experiments of the new paradigm.the average target localization accuracy reached 88.77%,among which one subject achieved96.52% of global localization accuracy.The high-resolution SSVEP-BCI is well adapted to multi-robot foraging tasks and enables human operators to grasp global range information,thus constituting a human-computer shared control framework with higher fault tolerance and sharing degree. |
