Self-paced EEG-based Brain-controlled car in Real-World Enviroment

Yang Yu, Zongtan Zhou, Jun Jiang, Yadong Liu, Dewen HuAdvances 2015 · DOI

Abstract

This study presented the design and online experiments of a self-paced EEG-based BCI for controlling a car in real world environment. This paradigm using two distinct MI tasks to generate a multi-task car control strategy, including start, move forwards, turn left, turn right, move backwards, and stop. The experiment results suggested that a brain-actuated car is possible when it runs in low velocity, approximately 5km/hour in our study. This proposed self-paced EEG-based BCI paradigm could potentially help disabled and paralyzed people to gain more mobility in the future, and moreover, provide a supplementary car driving strategy to healthy people.

Introduction

Transitioning Brain-Computer Interfaces (BCIs) from controlled laboratory settings to real-world applications remains a significant challenge. Most non-invasive BCIs, such as those based on Electroencephalography (EEG), suffer from low information transfer rates and typically require synchronous, computer-cued timing. In a driving scenario, a "self-paced" or asynchronous system is essential, allowing the user to decide when to issue a command and when to remain in a neutral state. This research addresses these hurdles by mapping a limited set of mental tasks—imagining left- versus right-hand movement—to a comprehensive suite of vehicle controls.

Methodology

The system architecture comprises an EEG processing unit, a user interface, and a vehicle control system interfaced via a CAN bus. The core innovation is a hierarchical, two-stage asynchronous classification strategy.

First, a Control Intention Classifier (CIC) distinguishes between a "Rest-state" and a "Motor Imagery (MI)-state." This ensures the vehicle does not react to idle thoughts. Once an intention is detected, a Left/Right Classifier (LRC) identifies whether the user is imagining left- or right-hand movement.

To expand these two binary mental states into six commands (Start, Forward, Left, Right, Backward, Stop), the researchers implemented a protocol tree. This logic allows users to navigate a command menu through sequences of MI tasks.

The protocol tree translates binary motor imagery tasks into six distinct driving commands.

Experimental Setup

Four healthy male participants (ages 22–26) participated in the study. EEG data were recorded using 16 electrodes targeting the sensorimotor cortex. The system was calibrated offline using Common Spatial Pattern (CSP) for spatial filtering and Linear Discriminant Analysis (LDA) for classification before proceeding to real-world online testing.

Results and Discussion

The system demonstrated robust performance in navigating a full-sized vehicle at a controlled speed of 5 km/h. Offline classification accuracy for the left/right tasks ranged from 71.3% to 89.8%, with a peak Information Transfer Rate (ITR) of 15.21 bits/min.

In online trials, participants successfully executed complex maneuvers, such as the "A-B-C" route involving both reversing and forward driving. One subject completed this task in just 8.12 minutes using only 14 commands, indicating high efficiency and minimal command errors.

The A-B-C navigation route used to test the system's real-world performance.

The "self-paced" nature of the BCI is its most critical feature, allowing for the distinction between active control and a resting state. While the current speed is low, the proof-of-concept provides a foundation for future mobility aids for individuals with severe motor impairments.

Limitations and Future Work

Current limitations include the low operating speed and the high mental demand required for sustained motor imagery. Future research will focus on increasing system reliability to support higher velocities, exploring "shared control" mechanisms where vehicle sensors assist in safety, and testing the system with the target demographic of paralyzed individuals.

Summary updated

Abstract

This study presented the design and online experiments of a self-paced EEG-based BCI for controlling a car in real world environment. This paradigm using two distinct MI tasks to generate a multi-task car control strategy, including start, move forwards, turn left, turn right, move backwards, and stop. The experiment results suggested that a brain-actuated car is possible when it runs in low velocity, approximately 5km/hour in our study. This proposed self-paced EEG-based BCI paradigm could potentially help disabled and paralyzed people to gain more mobility in the future, and moreover, provide a supplementary car driving strategy to healthy people.
BibTeX
@inproceedings{yu2015self,
  author = {Yang Yu and Zongtan Zhou and Jun Jiang and Yadong Liu and Dewen Hu},
  title = {{Self-paced EEG-based Brain-controlled car in Real-World Enviroment}},
  booktitle = {Advances},
  year = {2015},
  doi = {10.2991/icmmita-15.2015.258},
}
Figure