BROWSE
Detail
Conference Paper
System Overview for Multi-Task and Multi-Agent Deep Reinforcement Learning in Real Robot Environments
Cited - time in 
Share 
Share
- Authors
- Samyeul Noh, Hyonyoung Han, Chanwon Park, Junhee Park
- Issue Date
- 2020-10
- Citation
- International Conference on Control, Automation and Systems (ICCAS) 2020, pp.1239-1240
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- This paper presents an overview of the system for multi-task and multi-agent deep reinforcement learning in real robot environments. The presented system consists of three intelligence modules, such as cognition intelligence, task intelligence, and behavior intelligence, and is implemented in Python and the robot operating system called ROS. The cognition intelligence module utilizes a 3D vision camera to recognize all objects lying on the workspace by extracting useful features of objects. The task intelligence module receives a task from a user or a manufacturing process, builds a knowledge-based database for products, and assigns the task to the associated agents in a systematic manner. The behavior intelligence module determines the motion of each agent by pre-trained deep reinforcement learning algorithms or motion planning algorithms. This paper focuses on the system overview, including hardware configuration and software architecture.
- KSP Keywords
- 3D Vision, Deep reinforcement learning, Hardware configuration, Knowledge-based, Manufacturing processes, Motion Planning, Planning algorithm, Reinforcement Learning(RL), Robot operating system(ROS), learning algorithms, multi-agent