動的再構成可能ロボットシステムに関する研究 : 第24報, 隠れマルコフモデルを用いたDual-Peg-in-Hole作業のスキルの抽出

概要

論文の詳細を見る
A skill learning method is proposed in this paper. For a Dual-peg-in-hole problem, it is very difficult to analyze force and contact status. To extract the human skill of this kind of work, we record the force/torque data, sequence data, and psychological data by observing human work. Since human performance is inherently stochastic, we use a Hidden Markov Model (HMM) to represent these characteristics. By trainning the HMM using human data, we obtain a reasonable sequence of Dual-peg-in-hole work and the threshold of contact-state transition. We represent the human skill as a set of simple production rules that can easily be adopted into an actual manipulator system. This method is used in the assembly of self-organizing manipulator cells. The simulations and experiments illustrate the effectiveness of the algorithm.
1996-05-25