Four-Wheel Vehicle Driving by using a Spatio-Temporal Characterization of the P300 Brain Potential

Abstract

In this work a P300-based Brain Computer Interface (BCI) for the remote control of a four wheels vehicle, is presented. The proposed BCI exploits the P300 signal, an event-related potential (ERP) typically induced by visual/audio oddball paradigm stimulation protocol. For the driving purpose, in our application a four-choice synchronous BCI has been implemented. The neural interface architecture is made up by (i) the acquisition unit, (ii) the processing unit and (iii) the navigation unit. The former unit collects brain signals by 6 smart wireless electrodes from the parietal-cortex area. The processing unit is composed of a dedicated µPC (Raspberry Pi, RPi) performing stimuli delivery, the machine learning (ML) stage and the real-time classification. Specifically, the processing unit bases its ML stage working on a typical classification problem approach (i.e., feature extraction and classification). In this context, the main contribution of the work lies in the introduction of a P300 spatio-temporal characterization approach (t-RIDE), which allows to analyze all the available choices in a one-vs-all discrimination scenarios. It permits the implementation of very common binary classifiers despite the hyper dimensionality of the classification problem. Finally, the RPi-based navigation unit actuates the received commands and supports the vehicle by using peripheral sensors. As a proof of concept, the BCI operation has been tested on 7 subjects (aged 26 ± 3), using an acrylic prototype car. The experimental results showed that in the online free-drive mode (testing set), the BCI accuracy reached 84.28 ± 0.87% all over 4 choices, on single-trials.