バラ積みされた対象物のピッキングに対する物理シミュレータを用いた学習

DOI Web Site

書誌事項

タイトル別名
  • Learning Based Randomized Bin-picking Trained with Physics Simulator

抄録

<p>In this research, we tackle the problem of grasping objects randomly placed in a bin. Since complex physical phenomena in bin-picking make it difficult to predict the success or failure of picking, we consider introducing the deep learning. To obtain learning data, we use physical simulation where approximation is introduced in its collision checking to acquire data efficiently. In this paper, we first formulate the learning problems of bin-picking using CNN (Convolutional Neural Network). Next, we show that prediction of the success or failure of picking and derivation of optimum grasping posture are possible by using learned CNN. Finally, we indicate that the effect of approximation is relaxed when predicting the success or failure of picking.</p>

収録刊行物

キーワード

詳細情報 詳細情報について

問題の指摘

ページトップへ