Brain-controlled operator model-driven deep reinforcement learning for adaptive brain-machine collaborative control

The brain-controlled robot, a fusion of brain-computer interface and mobile robotics technologies, demonstrates enormous potential for applications across a wide range of fields. However, limited by the current state of EEG decoding technology, the brain-controlled process is often time-consuming and labor-intensive. For complex tasks, auxiliary control algorithms are required to build a brain-machine collaborative control framework that enhances the safety and autonomy of brain-controlled dynamic systems. While traditional auxiliary control algorithms can improve the safety to some extent, robot autonomy remains insufficient. To overcome this limitation, this paper proposes a Deep Reinforcement Learning-based Adaptive Brain-machine Collaborative Control (DRL-ABCC) method, which integrates a dual actor-critic mechanism for adaptive coordination between human and autonomous control under low EEG accuracy. In our method, a brain-controlled operator model (BCOM) is first constructed to simulate the decision-making of a human operator, generate brain-controlled commands, and integrate them into the training process of reinforcement learning. By embedding human decision-making experience into the training stage, the learning efficiency of the policy network is significantly improved. Furthermore, reinforcement learning uses its powerful autonomous exploration capability to enhance the system’s decision-making and adaptation capabilities in unknown environments, thereby improving task performance and autonomy. Finally, experimental results verify that the proposed DRL-ABCC method outperforms other baseline methods in the robotic target exploration task. Compared with other baseline methods, it significantly enhances the robot’s autonomy and robustness, especially in the case of low EEG accuracy, and achieves higher training efficiency.