Lateral Control of Brain-Controlled Vehicle Based on SVM Probability Output Model

The non-stationary characteristics of EEG signal and the individual differences of brain-computer interfaces (BCIs) lead to poor performance in the control process of the brain-controlled vehicles (BCVs). In this paper, by combining steady-state visual evoked potential (SSVEP) interactive interface, brain instructions generation module and vehicle lateral control module, a probabilistic output model based on support vector machine (SVM) is proposed for BCV lateral control to improve the driving performance. Firstly, a filter bank common spatial pattern (FBCSP) algorithm is introduced into the brain instructions generation module, which can improve the off-line decoding performance. Secondly, a sigmod-fitting SVM (SF-SVM) is trained based on the sigmod-fitting method and the lateral control module is developed, which can produce all commands in the form of probability instead of specific single command. Finally, a pre-experiment and two road-keeping experiments are conducted. In the pre-experiment, the experiment results show that, the average highest off-line accuracy among subjects is 95.64%, while for those in the online stage, the average accuracy is only 84.44%. In the road-keeping experiments, the task completion rate in the two designed scenes increased by 25.6% and 20%, respectively.