2P1-D13 人間共存ロボットの觝触適応行動 : 第23報:觝触特徴量の時間遷移に応じた觝触状態の連続的同定手法の提案

概要

論文の詳細を見る
In this paper, we propose a design method for constructing a continuous human-robot contact identification system based on transitional contact feature data. In the learning phase, our proposed system acquires a relationship, using a neural network called MCP, between verbalized contact states and contact features of tactile stimulation that allows distinguishing contact states such as 'stroke' and 'push', etc. In the cognition phase (where new contacts occur), transitional identification probabilities among contact states are provided every sampling time after matching in real time transitional contact feature data stored in the MCP in advance with those newly extracted in a constant duration traced back from any time. Experiments indicating that contact identifications were continuously gained every sampling time (0.02[s]) even if a contact is continuing or contact states discretely change such as from 'stroking' to 'pushing' confirm the effectiveness of the proposed method.
一般社団法人日本機械学会の論文