2P2-G01 実環境競合解消のための通信量を抑えた機械学習手法 : 複数台移動ロボット群における衝突回避に関するデッドロックの解消(進化・学習とロボティクス)

概要

論文の詳細を見る
Robot systems are gradually becoming more common in order to perform tasks that people can't. A multiple robot system is very difficult to work because the environment is very uncertain and information is often incomplete. Collision avoidance is a key problem, as autonomous robots must not collide with anything. There is pressure to use less and cheaper, lower quality sensors for widely spreading. Multiple robot systems will use collision avoidance algorithms which are based on low quality data, to in support of this situation. In this research, we designed a collision avoidance algorithm which use reinforcement learning to allow the robots to learn the most effective strategy. Computer simulation experiments were conducted in varying settings. We also experimented with real robots: autonomous lawn mowers RL500, and similarly good results were obtained.
2008-06-06