• 2P332   
•  

Controlling a Robot with Tactile Brain-computer Interfaces

•  
• Tomasz M Rutkowski:1,2　Kensuke Shimizu:1　Takumi Kodama:1　Peter Jurica:2　Andrzej Cichocki:2　Hiroyuki Shinoda:3　
• 1:Life Science Center of TARA, University of Tsukuba, Tsukuba, Japan　2:RIKEN Brain Science Institute, Wako-shi, Japan　3:The University of Tokyo, Tokyo, Japan　

We present the recently developed by our team tactile brain-computer interface paradigms applied for online control of robotic devices. The presented results have been obtained from online BCI-based robot control experiments with airborne ultrasonic tactile display (autdBCI), tactile pressure (tpBCI) and body tactile (btBCI) paradigms. Brainwave responses in the all presented online BCIs, tested with five healthy subjects, are captured with eight active EEG electrodes g.LADYbird connected to g.USBamp portable amplifier by g.tec, Austria, with 512 Hz sampling frequency. The experiments are conducted in oddball style paradigm eliciting the P300 responses classified next with SWLDA method and not exceeding four step averaging procedure in online experiments. The autdBCI experiments are conducted with a robotic hand controlled online with six commands. The tpBCI and btBCI paradigms are evaluated with the NAO humanoid robot walk control also based on the six pre-programmed movements.
[Method] The autdBCI stimulus generator produced vibrotactile contact-less stimulation of the human skin via the air using focused ultrasound. The effect was achieved by generating an ultrasonic radiation static force produced by intense sound pressure amplitude (a nonlinear acoustic phenomenon). The tpBCI tactile stimuli were delivered as light pressure patterns. Each tactile stimulus was generated via the tactile pressure generator composed of nine solenoids arranged in the 3 by 3 matrix in form of linear patterns. The tBCI somatosensory stimulation was delivered to larger areas of the user's back. The stimulated back areas were both shoulders, the waist and the hips using a haptic mattress developed by our team for the btBCI.
[Results] The BCI training sessions, based on six digits spelling (chance level of 16.6%) resulted with mean accuracies ranging from 83.5% to 100% for the all tested paradigms. The so trained SWLDA classifiers were next used in online experiments with robot control allowing for the final perfect accuracies with five trials based P300 responses averaging. The single P300 trial BCI experiments were not tested due to the longer command execution times required by the robots. These case studies demonstrate results obtained with the novel autdBCI, tpBCI and btBCI paradigms which has shown an interesting possibility to control on real time the robotic devices. The demonstration videos presenting the experiments are available online at http://bci-lab.info/demos/

研究助成：Research funds : The Strategic Information and Communications R&D Promotion Program, no. 121803027, of The Ministry of Internal Affairs and Commu