CASIA OpenIR  > 毕业生  > 硕士学位论文
机器人轴孔柔顺装配策略学习研究
王宇辰
Subtype硕士
Thesis Advisor王鹏
2019-05-28
Degree Grantor中国科学院自动化研究所
Place of Conferral中国科学院自动化研究所智能化大厦第七会议室
Degree Name工程硕士
Degree Discipline控制工程
Keyword机器人,轴孔装配,强化学习,ros
Abstract

近来由于机器人技术的不断发展,机器人的精度与执行效率都有了显著的提高,在现代工业发挥了越来越重要的作用装配是现代工业生产中的关键环节之一,将机器人应用于自动装配领域,替代重复性的装配动作,提高了生产效率、减少了劳动力消耗,在汽车制造、航空航天等领域获得了极大的成功。近年来,随着制造业的进一步快速发展,由于装配对象的多样性、装配环境和装配工艺的复杂性,给机器人装配的柔顺性和智能性提出了更高的要求,亟待研究具有一定适应能力的装配策略学习算法和系统。

本文针对典型的装配问题—轴孔装配问题,从实际接触力预测机器人自主学习两方面提出了不同算法,并搭建了一整套完整的仿真实际实验平台,以解决机器人主动柔顺装配问题。主要贡献有如下三个:

  1. 构建了机器人轴孔柔顺装配的仿真系统及实际系统:设计了基于机器人操作系统Robot Operating System, ROS的仿真实验平台能够快速迭代策略学习算法,令机器人能够自主学习柔顺装配策略。大长径比的轴孔装配问题作为实验对象设计了机器人轴孔柔顺装配的实际系统,为提出的柔顺装配算法提供了验证平台。
  2. 提出了一种新颖的面向机器人轴孔柔顺装配的接触力/力矩预测与分析模型提出了接触力/力矩预测模型,用于获取装配控制中至关重要的精确实际接触力及力矩;建立了接触状态分析模型,用于估计轴孔间的接触状态;最后,基于所提出的接触力/力矩预测和分析模型,设计了一种装配机器人位姿调整策略;通过在实际装配实验平台的实验结果,结果验证该算法能够满足大长径比轴孔柔顺装配的要求,模型预测的力/力矩误差小于1%,装配过程中平均力/矩小于5 N / 0.5 N·m
  3. 提出了一种基于动作-反向动作记忆的装配策略快速自学习算法基于强化学习原理对算法作出元素定义,并提出了一种新颖的动作-反向动作记忆模块,可以有效地减少机器人在接触状态空间的冗余探索,以此提高装配策略的学习速度;最后结合经典强化学习算法设计了装配策略快速自学习算法,在仿真实验系统上进行轴孔装配仿真实验,仿真结果验证该算法具有良好的力控性及拓展性
Other Abstract

Recently, the accuracy and execution efficiency of the robot have been improved significantly with the development of robotics, which plays an increasingly important role in modern industry. Assembly is one of the key links in modern industrial production. Robots are applied in the field of automatic assembly to replace repetitive assembly actions and improve production efficiency. They can reduce labor consumption and have achieved great success in automobile manufacturing, aerospace and other fields. In recent years, with the further rapid development of manufacturing industry, higher requirements have been put forward for the flexibility and intelligence of robot assembly due to the diversity of assembly objects and assembly environment and complexity of assembly process. Therefore, it is urgent to study the assembly strategy learning algorithm and system with certain adaptability.

This paper aimed at peg-in-hole assembly problem, which is the typical assembly problem, and proposed the algorithms based on the actual contact force prediction and robot autonomous learning. Moreover, the paper proposed a simulation system and a practical implementation system to solve the problem of active compliant assembly. The paper has three main contributions:

1. The simulation and practical systems of robot peg-in-hole compliant assembly were proposed. A simulation experiment platform based on Robot Operating System (ROS) was designed, which can quickly iterate the strategy learning algorithm, so that the robot can learn the compliant assembly strategy autonomously. A practical assembly system was designed for large length-diameter ratio peg-in-hole assembly problem, which provided a verification platform for the proposed compliant assembly algorithm.

2. A novel contact force/torque prediction and analysis model for flexible assembly of robot peg-in-hole assembly was proposed. We established a novel force/torque prediction model with measured data to obtain the precision actual contact force/torque which is critical for assembly control. A contact analysis model was built to estimate the assembly contact states. At last, based on the proposed contact force/torque prediction and analysis model, the paper designed a robot pose adjustment strategy. Through the experimental results in the actual assembly experimental platform, the results demonstrated that the proposed algorithm can meet the requirements of large length-diameter ratio peg-in-hole assembly. The force/torque error predicted by the model is less than 1%, and the average force/torque in the assembly process is less than 5N / 0.5N·m.

3. An assembly strategy learning algorithm based on action-reverse action memory was proposed. Firstly, the elements of the algorithm based on the reinforcement learning principle were defined in this paper. A novel action-reverse action memory method was proposed, which can effectively reduce the redundant exploration in the contact state space and improve the learning speed of assembly strategy. Finally, the paper proposed the assembly strategy fast learning algorithm based on the classic reinforcement learning algorithm. The peg-in-hole assembly experiments were completed on the simulation experiment system. The simulation results demonstrated that the learned strategy can be extended and control contact force well.

Pages83
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/23855
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
王宇辰. 机器人轴孔柔顺装配策略学习研究[D]. 中国科学院自动化研究所智能化大厦第七会议室. 中国科学院自动化研究所,2019.
Files in This Item:
File Name/Size DocType Version Access License
王宇辰_毕业论文_答辩前0517.pdf(2791KB)学位论文 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[王宇辰]'s Articles
Baidu academic
Similar articles in Baidu academic
[王宇辰]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[王宇辰]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.