Detecting EEG correlates during preparation of complex driving maneuvers

The main focus of this research is the development of real-time recording and analysis devices and algorithms for detecting Electroencephalography (EEG) correlates of steering actions. First, offline analysis where performed in a high fidelity driving simulator task using a commercial EEG system. The results of this study have been then confirmed with real world recordings. Finally, online algorithm has been tested. The participants have driven on a simulation of a real highway between Geneva and Lausanne (A1) perform-ing self-paced lane changes. EEG has been collected together with steering actions. Analysis of motor cortex activity preceding the onset of steering movements has confirmed the presence of Motion Related Cortical Potentials (MRCP). Furthermore, a systematic evaluation has identified parameters that maximize the detection with minimal number of sensors. Using 9 sensors over the motor cortex has been sufficient to generate a single-trial classification True Positive Rate (TPR) of 80.2% at about 420 ms before the onset of the movement. These results have further guided the development of a new wireless, dry-sensor EEG recording system dedicated for car applications. The newly developed system has been tested and proved to be less intrusive and to provide signals of similar quality. Recordings have continued on the real world, the actual A1 highway between Geneva and Lausanne. In these experiments, EEG has been collected together with vehicle control data. The participants have driven an Infiniti FX30 vehicle at constant speed with Intelligent Cruise Control support and have been asked to per-form lane changes in as many situations as possible. Not only that the presence of MRCP has been confirmed as preceding the onset of steering, but applying the machine learning algorithms developed in the Driving Simulator have yielded similar results, interestingly showing that the increased strength of the MRCP due to higher involvement and arousal levels compensate for the much noisier environment. Furthermore, the machine learning parameters tuning framework has been automatized and updated for online detection. Several classifiers that yield highest TPR peaks in consequent time windows are automatically chosen and a fusion step has been added. While participants have driven in the Driving Simulator, a set of classifiers has been trained on automatically detected lane changes trials and for the last part of the re-cording the output of the fusion has been evaluated. Impressively for such complex maneuvers like the steering for lane changes, a TPR peak of 70% at 370 ms before the onset of the movement has been con-firmed. While not being ready for engineering applications, the current research has laid a strong scientific foundation for real-time detection of EEG correlates of steering actions. The fact that the results have been con-firmed both in the laboratory and real world proves that developing BCI interfaces for intelligent vehicles is a