一种基于MADDPG的AGV动态避障方法
首发时间:2019-03-15
摘要:深度增强学习将深度学习的感知能力和增强学习的决策能力相结合,在智能控制,机器人控制及预测分析等领域有广泛应用空间。本文将AGV动态避障问题置于深度增强学习框架下,利用场景中物体位置和速度等可采集信息,采用MADDPG深度增强学习框架进行训练,利用经验回放机制和分散执行、集中评估的训练模式,以及设计合理的奖励函数,使得AGV在动态环境下有一定的自主避障能力。
For information in English, please click here
Application of MADDPG in Dynamic Obstacle Avoidance of AGV
Abstract:Deep reinforcement learning which combines the perception of deep learning with the decision-making ability of reinforcement learning, has a wide application in intelligent control, robot control and predictive analysis. The paper deal with the AGV Dynamic Obstacle Avoidance (DOA) problem under the Deep reinforcement learningframework. the robot information such as the position and velocity, and the MADDPG framework, experience replay and skillfully reward function was applied, AGV has ability to avoid obstacles in the dynamic environment.
Keywords: Intelligent system AGV DOA MADDPG
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
一种基于MADDPG的AGV动态避障方法
评论
全部评论0/1000