センサの見通し可能性を考慮した確率的モデルによる障害物地図の作成

概要

論文の詳細を見る
This paper describes an obstacle map generation by bayesian theorem. A sensor for the map generation can observe the obstacle which sensor can look, and cant observe the obstacle which is occluded from other obstacles. Previous method cant generate the actual map with the observation which has substantial margin of error because generate the map by distant observation model without considering occlusion. In our research, we use the observation model considering occlusion. In order to considering occlusion correctly, we estimate a visibility: Which obstacle is observed? can observe. Additionally in order to estimate the visibility, we have to consider spatial dependencies that the obstacle and the free area has size of space. In order to consider spatial dependencies, we estimate the visibility by a assumption that each line of view is markov chain. Experimental result shows that we obtained more precise generation of the obstacle map by our method than by previous method.
社団法人日本ロボット学会の論文