一种面向六自由度机械臂柔顺装配的深度强化学习策略
A flexible assembly strategy for a 6-DOF robotic arm based on deep reinforcement learning
张宇航 1陈雯柏 1张佳琪 1马航1
作者信息
- 1. 北京信息科技大学 自动化学院,北京 100192
- 折叠
摘要
针对现有3C装配方法存在装配复杂性高、柔性要求高和零部件多样性等问题,使用深度强化学习方法,利用MuJoCo物理引擎,对机械臂装配任务建模,搭建了机械臂3C装配系统,设计了该系统的状态空间和动作空间,通过奖励函数设计缓解了奖励稀疏的问题.仿真结果表明,所提出的装配系统高效和准确地完成了 3C装配任务,使用的SAC算法装配成功率达到93%,高于其他算法.同时,经过强化学习策略训练过后,装配接触力达到设定范围,实现柔顺装配,研究结果对提高生产效率和产品质量有着重要意义.
Abstract
To address the problems of the current 3C assembly methods,such as high assembly complexity,high flexibility and excessive parts,we build a 3C assembly system for robotic arm by employing a deep reinforcement learning method.First,the MuJoCo physics engine is used to model the assembly task of the robot arm and build a simulation environment.The task is modeled as a Markov process problem.The state space and action space of the system are designed and the problem of sparse reward is alleviated through the design of reward function.A series of simulation experiments show our proposed assembly system efficiently performs the 3C assembly task and achieves good accuracy and stability.Moreover,the assembly success rate reaches 93%.Meanwhile,after the reinforcement learning strategy training,the contact force reaches the set range and achieves flexible assembly,greatly improving the production efficiency and product quality.
关键词
深度强化学习/3C装配/机械臂控制Key words
deep reinforcement learning/3C assembly/manipulator control引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据