BMW: Brainwave Manipulated Wagon Zijian Chen, Tiffany Jao, Man Qin, Xueling Zhao Faculty Advisor: Prof. Qiangfei Xia Abstract System Overview BMW (Brainwave Manipulated Wagon) is a robotic car that can be remotely controlled by user’s brain EEG (Electroencephalography) signals. Our system uses BCI (Brainwave Computer Interface) to provide the communication between our brain, the computer application and the robotic car. The system uses a commercial EEG headset to acquire EEG data. Then, the data is processed and classified into command in the computer application. The application transmits the command signal to the robotic car, which will achieve desired operation. User is able to control the car to move forward, backward, stop and turn left and right. Our system primarily utilizes two responses in the occipital lobe region of the brain, which is more responsive to visual perception. I. Eye-Closed i. Results in increases in alpha wave power ii. Applies Bayesian classifier for the eye-closed detection II. SSVEP (Steady-State Visual Evoked Potentials) i. Invokes by focusing on a light stimuli blinking at fixed frequency ii. Observes consecutive detection of dominant frequency Background I. Eye-Open/ Eye-Closed wireless 2.4GHz (real time) dominant frequency SSVEP Classifier Eye-Open EEG Voltage Data vs Time 4420 Voltage(uV) 4400 Voltage(uV) 4380 4360 4340 4320 4300 1 2 3 40 30 30 20 10 0 4 6 8 10121416 22 Database (Training Data) (training) Alpha, beta power command 4 5 283032 40 45 50 20 10 0 01 6 8 10 12 14 16 22 28 30 32 40 45 50 Frequency(Hz) Figure(d) II. SSVEP Arduino • Bluetooth HC05 Robotic Car Control the car to turn left or right upon detection of SSVEP signal. Magnitude vs Frequency for 10Hz Stimuli Magnitude vs Frequency for 15Hz Stimuli 1000 1000 800 800 Magnitude User Action Avg. Detection Accuracy Time (sec) Magnitude Specifications 600 400 200 600 400 200 0 0 0 Close eyes Forward if the car is currently stopped Close eyes if the car is current moving forward/backward Left Turn Stare at 10Hz light stimulus Right Turn Stare at 15Hz light stimulus Backward 3 Figure (a) and (b): Voltage EEG Data versus Time. Voltage EEG data is the voltage measurement resulting from neuron activities. These two graphs do not show much information about alpha and beta wave. Figure (c) and (d): Now, the data is interpreted in frequency domain. A dominant spike is observed within the alpha wave region under eye-closed condition, which causes the total alpha wave power to increase significantly. At the same time, beta wave power does not change much. Emotiv EPOC headset Stop 2 Eye-Closed FFT Magnitude vs Frequency Frequency(Hz) Figure(c) User Interface Command 1 Time(Sec) Figure(b) 40 (training)Alpha, beta power Bayesian Classifier 5 Eye-Closed Raw voltage EEG data vs Time Eye-Open FFT Magnitude vs Frequency Computer (real time) Alpha, beta power 4 4420 4400 4380 4360 4340 4320 4300 Time(sec) Figure(a) 0 C# Application Signal Processing Control the car to move forward, backward, and stop by detecting the rise of alpha wave power. Magnitude Block Diagram EEG Voltage data • Magnitude Brainwaves are produced by synchronized electrical pulses from masses of neurons communicating with each other. Types of Brainwaves: – Delta 0-4Hz – Theta 5-7Hz – Alpha* 8-12Hz – Beta 13-30Hz – Gamma 30+Hz *Occipital alpha waves during eye-closed periods are the strongest EEG brain signals. Close eye for a significant longer time 5 10 15 20 25 Frequency(Hz) Figure (e) 2.15 7.04 90.9% 86.7% 6.51 90.5% 6.01 77.3% 0 5 10 15 20 25 Frequency(Hz) Figure (f) Figure (e): A significant spike at 9.25Hz can be observed when staring at the 10Hz stimuli Figure (f): A significant spike at 12.75Hz is observed when staring at the 15 Hz stimuli Acknowledgement Specially thank Prof. Xia for being our advisor, providing us feedback and sponsoring us to purchase the Emotiv EPOC headset. Thank you to Prof. Soules, Prof. Tessier, Prof. Mettu of Tulane University, Mr. Alexander de Geofroy, Prof. Rebecca Spencer and the Cognac Lab. We also want to thank Fran Caron and Terry Bernard for ordering the parts for our project. Department of Electrical and Computer Engineering ECE 415/ECE 416 – SENIOR DESIGN PROJECT 2015 College of Engineering - University of Massachusetts Amherst SDP15 User Interface Eye-Closed Detection The eye-closed detection is based on the Bayesian classifier. 1. Find Reference Point: Determines the threshold for high alpha and high beta based on the training set data 2. Analysis Training Set: Updates the probability table to calculate the classified result 3. Classification: Determines if the real-time alpha and beta power are more likely to be eye-closed or eye-open Training Interface • Gives user the direction to perform the corresponding training • Gathers alpha and beta power during the eye-open and eye-closed training Sensor Contact Form • Checks if the reference sensors and the O1 sensors are in good contact Forward/Stop/Backward Close eyes until the car reacts/detect alpha high Car Forward Keep eyes closed after car stops until car backward Car Backward Car Stop Close eyes until the car stops/detect alpha high Close eyes until the car reacts/detect alpha high Control Interface • Display – 10 Hz and 15 Hz light stimuli for invoking SSVEP response – The real-time command detection result – Alpha and Beta Power vs Time graph • Control – Connection with the robotic car – Start / Stop the command detection SSVEP Detection Results 1. Analysis Training Set: Determines the dominant frequency when the user is looking at the 10 Hz and 15 Hz blinking square 2. Classification: Checks for consecutive occurrence of dominant frequency Experimental Results: – 9.25 Hz for left turn – 12.75 Hz for right turn Patten Observation Source of Error EyeOpen/ EyeClosed Alpha power increases significantly compared to beta power SSVEP Dominant Frequency around: 1. Brightness of screen - 9.25 Hz for starting at the 2. Accidentally glances at 10Hz stimuli the wrong blinking - 12.75 Hz for staring at the square 15Hz stimuli 1. Muscle Movement 2. Tiredness - also leads to increase in alpha power Robotic Car Cost Development Production Part Price Part Price Arduino Uno 24.51 Arduino Uno 19.96 Motor * 2 3.90 Motor *2 3.90 Bluetooth receiver 7.59 Bluetooth receiver 7.59 L298N Motor Driver 34.95 L298N Motor Driver 27.96 EPOC Headset Total EPOC Headset Total 699 769.95 699 758.41 HC-05 Bluetooth module is used to receive data, which is a digital signal from the computer. An Arduino board is used to retrieve the number and convert it into four digital output. Output pins are hooked to L298N Motor driver board. L298N controls the two motors, which allows the car to move.