Brain-Controlled Wheeled Mobile Robots: A Framework Combining Probabilistic Brain-Computer Interface and Model Predictive Control

Abstract

Brain-controlled systems have experienced significant advancements in overall performance, largely driven by continuous optimization and innovation in electroencephalography (EEG) acquisition experimental paradigms and decoding algorithms. However, their applications still face challenges, including limited control precision and low efficiency. In this article, we focus on a wheeled mobile robot (WMR) as the control object and propose a novel brain-controlled framework that combines a probabilistic brain-computer interface (BCI) and a model predictive controller (MPC). First, the probabilistic BCI is developed, featuring the sigmoid fitting-filter bank canonical correlation analysis (SF-FBCCA) algorithm, which serves as the core of the BCI system by decoding EEG signals and generating brain commands along with their associated probabilities. Second, an auxiliary MPC is integrated into the probabilistic BCI system to provide decision-making assistance while preserving the users' primary brain control authority. The weights of the cost function are adaptively determined based on the command probabilities. Finally, simulation-based evaluations were conducted using the WMR in a path-keeping scenario. The results demonstrate that the proposed framework significantly improves control accuracy and efficiency compared to direct brain control approaches, reducing the average lateral error by 58.02% and the average yaw angle error by 60.06%. Additionally, the MPC employing adaptive weights further improves overall performance. These findings offer theoretical insights and technical references for future research on BCI-based control frameworks.