ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper A Scalable Client-server based Multi-drone Simulation Architecture Supporting Efficient Machine Learning
Cited 1 time in scopus Share share facebook twitter linkedin kakaostory
Authors
Soojeon Lee, Byoung-Sun Lee, Jaeyoung Ahn
Issue Date
2018-10
Citation
International Conference on Applied Computing (AC) 2018, pp.380-384
Publisher
IADIS
Language
English
Type
Conference Paper
Abstract
The proliferation of machine learning approaches has brought a body of interesting research topics such as deep learning and reinforcement learning. While enhancing the level of machine intelligence is the main focus in this area, another interest lies in how to easily speed up the learning process itself. However, for some domains such as drone autonomous flight, it is difficult to enable fast and efficient learning in real world since flying drones is unsafe, expensive and time-consuming. To cope with these limitations, some recent simulation-based approaches provide physically and especially visually realistic virtual environments and thus enable to collect a large amount of training data fast in various conditions. However, they do not scale well due to their centralized architecture working in stand-alone mode. Thus, various important scenarios where multiple drones coexist (e.g., formation flight) in a space cannot be simulated. In this paper, we introduce our architecture enabling scalable multi-drone simulations. Our architecture, which is based on a distributed client-server model is scalable and aims to make many drones flight and interact with others in the same space simultaneously.
KSP Keywords
Amount of training data, Autonomous flight, Centralized architecture, Client-Server model, Efficient learning, Learning Process, Machine Learning Approach, Real-world, Reinforcement learning(RL), Research topics, Simulation Architecture