2013 年 31 巻 1 号 p. 89-97
This paper deals with evaluating objective Falling Risk and locomotion selection algorithm based on Falling Risk and moving efficiency. A robot evaluates Falling Risk as an Indicator of uncertainty using Bayesian Network based on the error and environment information obtained from sensors. For objectivity of Falling Risk, we have a robot learn conditional probability table and Bayesian Network model by simulation. Locomotion selection model during walking is Semi‐Markov Decision Process, so a robot update locomotion reward that is based on moving velocity as moving efficiency and Falling Risk until the swing leg land. When the swing leg land, the locomotion, having the maximum locomotion reward, is selected by greedy algorithm. As a result, a robot realize the move of environment that is difficult to move with only single locomotion on maintaining the maximum moving efficiency.