不確かさを考慮した視覚と行動のプラニング

抄録

This paper describes a new framework of unified planning of vision and motion for a mobile robot under existence of uncertainty. In mobile robot planning in the real world, the uncertainty and the cost of visual recognition are important issues to be considered. When a robot recognizes the environment with vision, efficient views must be selected considering a trade-off between the cost of visual recognition (including the cost of motion for recognition) and the effect of information obtained by vision. We use a probabilistic model to represent uncertainties and apply statistical decision theory to constructing a unified plan of vision and motion. We propose a method of predicting sensor information and formulate the planning problem in a recurrence formula. We can justify this formulation since we use Bayes theorem to integrate sensor data from multiple views. As an example of uncertainty modeling, we construct a model of the uncertainty of stereo vision. We also analyze the planning problem using a simple example and conclude that the combination of dynamic programming (DP) and the hill-climbing method is useful for searching for optimal solutions. We apply our planner to several planning problems and show the validity of our approach. Future works are also described.