The use of electroencephalogram (EEG) data is common to develop brain-computer interface (BCI) applications. Analysis of EEG data in the oddball paradigm has revealed that some electrodes experience clearer manifestations of the P300 wave, giving a particular relevance to their position. For this study, we arrange recorded EEG data as a single trial 3D representation in which spatial and temporal information are kept with few to no modifications for off-line classification. This differs from most studies using EEG data and convolutional neural networks (CNN), for which a 2D-mapped version of the data is normally used. The recorded data used for this study corresponds to the EEG of four subjects that were presented with auditory stimuli via earphones from six different virtual directions following the oddball paradigm to elicit the P300 wave. The stimuli presentation was done at time intervals of 500, 400, and 300 ms to verify the impact of the speed in the performance of an auditory BCI. The results obtained for the identification of P300 within the EEG shows that the 3D input proposed in this study is a reliable approach for input configuration and that the CNN models trained using it tend to correctly differentiate the EEG containing the P300 from that which doesn't with low accuracy variations regardless of the differences of the used models.

In the study of brain computer interface using electroencephalogram (EEG), the area of motor imagery of individual finger movement has been extensively examined. The objective of this study is to predict which finger is moving or being imaged through the use of EEG. We measured EEG activity while subjects performed either motor execution or motor imagery of individual finger movements with their right hand. Event related spectral perturbation was used as indicator of brain activity, which represents frequency power fluctuation from the baseline interval. The frequency bands α, β, and γ (8-15, 16-31, and 32-128 Hz, respectively) were used for analysis. Additionally, a feature consisting of a combination of those three bands was also explored using principal component analysis. These four kinds of features were classified by a linear discriminant analysis using ten-fold cross validation. Result indicated that the combined feature of the three frequency bands could classify most combinations of finger in both motor execution and motor imagery. The results suggest that EEG during the performance of motor imagery of individual finger movement is discriminable depending on the subject.

Humans modify their movement characteristics according to the requirement of the movement accuracy. The relations between the required accuracy and muscle activities were previously determined by varying the size of the targets to be reached in isotonic contraction tasks. However, in isometric contraction tasks, such relations have not been studied yet. In this study, we examined whether there was a difference in the movement characteristics depending on the sizes of the targets being introduced. We observed that the distribution of the forces at the end of the movement varied with the target sizes, similar to results of previous studies based on isotonic contractions. Furthermore, we noted a trade-off relation between the required accuracy and co-contraction of the muscles. When we examined the muscle synergies, we observed that for some subjects, the number of synergies increased with the increase in target size. This result suggested that the strategy of motor control might be modified depending on the requirement of accuracy of the isometric reaching task considered herein.

Brain-Computer Interfaces (BCI), which are capable of operating external devices using human brain activity, have recently been actively studied. Previous studies have developed an auditory BCI using event-related potential (P300) for estimating the single target of a sound source direction to which the subjects pay attention. In this study, we proposed a new paradigm using two targets instead of a conventional one-target paradigm and examined whether it was possible to increase the information transfer rate using this two-target paradigm. We conducted experiments adopting proposed two-target paradigm where two out of six sound directions were defined as a target. Considering the conditions obtained more than 70 % of identification rate of the target direction of sound, results for two-target paradigm showed higher information transfer rates than that for a conventional one-target paradigm for three out of five subjects. From the above, increasing of the target direction might be an effective means to increase information transfer rate of auditory BCI.

Fitts' law is an empirically derived model that describes movement duration during human arm reaching as a function of movement distance and target width. Although this law does not consider dynamic properties, such as motor commands and joint torques, it can explain the speed-accuracy trade-off characteristics of human arm movements in most cases. However, human arm movements are achieved by transforming motor commands through dynamics. Thus, speed-accuracy trade-off characteristics can be explained by the joint torque produced by motor commands. Here, we investigated whether the speed-accuracy trade-off described by Fitts' law can be better explained by joint torque, which is directly controlled by motor commands. To examine the relationships between the movement speed, end-point accuracy, and joint torque, we performed temporally and spatially constrained experiments. A positive correlation was observed between the integrated absolute torque and end-point error, which is a measure of end-point accuracy for the trial-averaged values of each experimental condition. These findings suggest the possibility that the speed-accuracy trade-off described by Fitts' law can be explained by human arm dynamics and its parameters.

Muscle synergy is defined as a combination of the limited number of muscle activities, which has been considered useful for controlling a large number of degrees of freedom in the musculoskeletal system. In previous studies, the robustness of muscle synergy recruitment across biomechanical tasks has been reported, i.e., the central nervous system controls a few pre-organized muscle synergies corresponding to a specific motor behavior. In contrast, the present study considers a different hypothesis wherein muscle synergy recruitment can be affected by muscle contraction. In our experiment, the subjects were instructed to perform an additional muscle contraction during the eight directional point-to-point reaching movement, on the basis of visual feedback of four levels of contraction. The result of extracted muscle synergies from electromyogram across contraction levels that as the level of contraction increases, so does the number of recruited muscle synergy. Moreover, we found a high similarity between shared muscle synergies-the first synergy under different levels of contraction. The result revealed the hypothesis that pre-organized synergies may be responding to the specific motor task but this coordination of existing muscle synergies can be adjusted by muscle contraction. This adjustment can be useful to develop the more complex myoelectric interfaces efficiently which could control not only arm motion but also arm force.

The accurate detection of event-related potentials (ERPs) is of great importance to construct brain-machine interfaces (BMI) and constitutes a classification problem in which the appropriate selection of features from dense-array EEG signals and tuning of the classifier parameters are critical. In the present work, we propose a method for classifying single-trial ERPs using a combination of the Lifting Wavelet Transform (LWT), Support Vector Machines (SVM) and Particle Swarm Optimization (PSO). In particular, the LWT filters, the set of EEG channels and SVM parameters that maximize the classification accuracy are searched using PSO. We evaluate the method's performance through offline analyses on the datasets from the BCI Competitions II and III. The proposed method achieved in most cases a similar or higher classification accuracy than that achieved by other methods, and adapted wavelet basis functions and channel sets that match the time-frequency and spatial properties of the P300 ERP.

Localization of virtual sound source is a technology that allows the reproducing of three-dimensional sounds using stereo earphones, and applications of this technology in Brain-machine interface that use auditory stimuli are being investigated. In order to achieve virtual sounds using this technology, the Head-related Transfer Function (HRTF) of the user must be measured accurately. The HRTF can be measured accurately with the appropriate placement of the microphones and measurement environment, but procuring an ideal setup is usually difficult. To overcome this, we instead attempt to obtain an accurate HRTF using reinforcement learning. We performed simulations and verified that with the proposed method the HRTF accuracy improved on 24 horizontal directions. Also, in online learning experiments, the localization accuracy was improved for 3 subjects, suggesting the validity of our method.

A method is described for classifying near-infrared spectroscopy (NIRS) signals measured for motor imagery and/or execution using the left or right hand. The measurement time intervals and the signal channels are used as features. The signals are discriminated using a support vector machine. Experiments demonstrated that this method has a higher generalization capability than a previous method for classifying NIRS signals. Testing of its ability to classify the signals according to whether they are for right- or left-hand motor imagery and/or movement demonstrated that its classification of NIRS signals satisfies the two-category classification problem. A promising application is to brain-computer interfaces, a potential communication tool for paralyzed individuals.

We applied event-related potential (ERP) to reinforcement signals that are equivalent to reward and punishment signals. We conducted an experiment using an electroencephalogram (EEG) in which volunteers identified the success or failure of an inverted pendulum task. We confirmed that there were differences in the EEG signal depending on whether the task was successful or not and that ERP might be used as a punishment of reinforcement learning. We used a support vector machine (SVM) for recognizing the ERP. We selected the feature vector in SVM that was composed of averages of each 35 msec for each of three channels (F3,Fz,F4) on the frontal area, for a total of 700 msec. Our experimental results suggest that reinforcement learning using ERP can be performed accurately. Finally, we suggest the possibility of developing an adaptive brain-computer interface (BCI) by ERP.

As a novel learning method, reinforced learning by which a robot acquires control rules through trial and error has gotten a lot of attention. However, it is quite difficult for robots to acquire control rules by reinforcement learning in real space because many learning trials are needed to achieve the control rules; the robot itself may lose control, or there may be safety problems with the control objects. In this paper, we propose a method in which a robot in real space learns the virtual task; then the task is transferred from virtual to real space. The robot eventually acquires the task in a real environment. We show that a real robot can acquire a task in virtual space with an input device by an example of an inverted pendulum. Next, we verify that the acquired task in virtual space can be applied to a real world task. We emphasize the utilization of virtual space to effectively obtain the real world task.

Humans can generate a complex trajectory by imitative learning of others' movement. A method for learning complex sequential movements, but utilizing a via-point representation is proposed. However, the proposed algorithm for estimating a set of via-points from the complex movement does not involve a learning process such as a learning by trial and error. The algorithm can find the minimum number of via-points, and then specify the unique set of via-points without a trial-and-error process. In this paper, we report an acquisition algorithm for via-point representation through trial and error in a human-like manner. The proposed via-point acquisition algorithm based on reinforcement learning finds a set of via-points that can mimic the reference trajectory by iterative learning using evaluation values of generated movement pattern.