https://scholars.lib.ntu.edu.tw/handle/123456789/581381
標題: | Real-time obstacle avoidance using supervised recurrent neural network with automatic data collection and labeling | 作者: | Chan S.-H Xu X Wu P.-T Chiang M.-L LI-CHEN FU |
關鍵字: | Motion planning; Stars; Automatic data collection; Local path-planning; Occupancy grid map; Real time obstacle avoidance; Rule-based method; Simulated experiments; Supervised recurrent neural network; Training process; Recurrent neural networks | 公開日期: | 2019 | 卷: | 2019-October | 起(迄)頁: | 472-477 | 來源出版物: | Conference Proceedings - IEEE International Conference on Systems, Man and Cybernetics | 會議論文: | 2019 IEEE International Conference on Systems, Man and Cybernetics, SMC 2019 | 摘要: | In this paper, we propose an approach for real-time obstacle avoidance based on a supervised Recurrent Neural Network (RNN). As compared with conventional rule-based methods, fewer hyper parameters are needed to be tuned in the proposed system. On the other hand, as a data-driven system, our approach generates training data autonomously without manual labeling process. One of the main features of the proposed system is data generation, which can provide thousands of training data for supervised learning using simply 2D occupancy grid maps as input. To efficiently generate the path data, we utilize A{star } algorithm as the initial guide for the autonomous training process of the RNN model. After that, the trained model will perform local path planning to avoid obstacles, which is tested in practical environments. With the proposed approach, we can effectively reduce the training time while maintaining satisfactory performance. Simulated experiments show that the proposed system not only exhibits the features of A{star } algorithm in global aspect for path planning, but also performs obstacle avoidance in local aspect. As a by-product, the simulation results also show that the autonomously trained model can be successfully applied to many different scenarios. ? 2019 IEEE. |
URI: | https://www.scopus.com/inward/record.uri?eid=2-s2.0-85076743126&doi=10.1109%2fSMC.2019.8914281&partnerID=40&md5=a66b7c499f6e0761acb0a9a179376e68 https://scholars.lib.ntu.edu.tw/handle/123456789/581381 |
ISSN: | 1062922X | DOI: | 10.1109/SMC.2019.8914281 |
顯示於: | 資訊工程學系 |
在 IR 系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。