成果報告書詳細
管理番号20160000000582
タイトル*平成27年度中間年報 次世代ロボット中核技術開発 (革新的ロボット要素技術分野)ブレイン・マシン・インターフェース 脳活動モデル同定と内部状態推定に基づくBMI技術 
公開日2016/7/22
報告書年度2015 - 2015
委託先名株式会社国際電気通信基礎技術研究所
プロジェクト番号P15009
部署名ロボット・AI部
和文要約
英文要約Title:Strategic Advancement of Multi-Purpose Ultra-Human Robot and Artificial Intelligence Technologies (Future robot technology) Constructing a BMI though identifying dynamics of brain activities and estimating internal brain states (FY2015-FY2016) FY2015 Annual Report

(1) Objectives
Electroencephalography (EEG)-based Brain-Machine Interface (BMI) has been explored to be used as an interface between the brain and a robot. However, because of the low signal-to-noise ratio, it takes a long time to make a BMI decoder estimate a correct user’s intention, limiting the practicality of such interface. This project aims at developing a machine learning (ML)-based approach to estimate the user’s internal state from non-stationary and noisy EEG signal.
(2) Outline
Traditionally, an accurate estimation of a user’s intention requires more than 5 sec EEG data due to the low signal-to-noise ratio. This decoding delay limits practical application of EEG-based BMI on robots. We propose to predict the internal state of the signal source to compensate the delay, using ML-based technology. By applying a successive state prediction approach, which combines the user’s motion intention induced by a robot’s feedback with an observed EEG signal, a BMI output is produced. Wearable robots are used to realize a brain-robot closed loop system, so that the brain and the BMI system mutually adapt to each other through their interaction. We attempt to define a dynamic model of the brain activity and to deploy the “zero-delay” brain-based robot controller in a real application.
(3) FY2015 Report
Goals of FY2015 are to survey technical trends on the estimation of the user’s control intention and on the construction of a brain-robot closed loop system with adaptive BMI.
(3-1) Brain activity
Among the available types of brain activity, sensorimotor rhythms (SMR) are considered as one of the most suitable strategies for the natural control of a neuroprosthesis, since they originate in cortical areas directly connected to the normal neuromuscular pathways. Moreover, being SMR an endogenous signal, it can be used to operate a BMI autonomously, without the need of external stimuli.
(3-2) BMI Robot control
Recent studies have been testing the hypothesis that the haptic neuro-feedback provided by the robot could promote motor recovery in stroke patients. Most of these works focused on motor imagery (MI), given that the reinforcement of sensorimotor cortex activation by haptic neuro-feedback is the closest strategy to active training by motor execution. We observe that, the majority of the studies have been addressing the upper limbs, which is mainly due to the difficulties associated with the control of biped walking and balancing of a lower limb exoskeleton robot.
(3-3) Suitable BMI paradigms
Typical SMR-based systems suffer from non-stationarity and BMI `illiteracy'. With respect to the first problem, adaptive classifiers can mitigate the effect of actuations in the feature space. Co-adaptive BMI paradigms have been proposed to address `illiteracy' while minimizing calibration time. In order to evaluate the suitability of the aforementioned BMI paradigms, we will implement and test four different SMR-based decoders in the next FY2016
ダウンロード成果報告書データベース(ユーザ登録必須)から、ダウンロードしてください。

▲トップに戻る