CASIA OpenIR  > 毕业生  > 硕士学位论文
面向运动目标跟踪的机械臂补偿学习控制技术研究
王丽丽
2023-05-25
Pages92
Subtype硕士
Abstract

视觉是机器人获取外界信息的重要途径,基于视觉反馈的机器人运动控制 在机器人领域中具有重要的地位。随着机器人应用领域的不断扩展,需要机器人 操作运动目标的场景越来越多,如抓取运动的物体、装配移动的轴孔等。使用视 觉引导机器人跟踪并操作运动目标成为一项研究热点。

在机器人操作静态目标的视觉控制技术中,通常忽略指令传输和图像处理 等造成的时延。但在操作运动目标时,时延会降低跟踪性能甚至导致操作失败。 此外,未知或不确定的系统参数也会影响跟踪性能。因此本文针对视觉控制跟踪 系统中模型不确定性、时间延迟和其他干扰因素造成的跟踪性能下降问题,基于 机器人跟踪系统的当前状态和历史状态优化机器人跟踪控制器参数,实现高精 度的运动目标跟踪并可靠操作目标。本文的主要工作和贡献归纳如下:

1. 提出一种基于延时补偿的视觉反馈自适应控制系统 面向“眼在手上”机器人跟踪系统及运动轴孔装配任务,提出了一种基于图 像伺服的机器人视觉跟踪方法。为解决不确定的视觉伺服参数导致的机器人跟 踪精度下降问题,提出伪逆图像雅可比矩阵自适应算法来在线调整图像雅可比 矩阵。此外,构建了基于径向基神经网络的时间延迟补偿器来消除闭环跟踪系统 中指令传输和图像处理延迟带来的影响,从而减少由时滞引起的跟踪误差。基于 延时补偿的视觉反馈自适应控制系统包含了伪逆图像雅可比矩阵自适应和时滞 补偿两部分。本文通过仿真和实体实验验证了所提方法的有效性,并成功完成了 轴孔间隙仅为1cm且处于运动状态下的轴孔装配任务。

2. 提出一种基于强化学习补偿的视觉反馈控制系统 针对三维空间中任意运动目标的跟踪问题,设计了一种基于滑模控制器和 SiamBAN网络的视觉跟踪控制器。构建了基于Actor-Critic强化学习框架的伺服 控制扰动补偿模型,以补偿传统伺服控制器无法消除的外部扰动、模型误差等造 成的跟踪误差。分别提出了控制输入补偿和参考补偿两种补偿器来增强基于滑 模控制器的视觉伺服性能。所提出的方法结合了传统机器人控制与强化学习的 优势,通过自调节的方式在线优化提升跟踪性能,并保证了机器人运行过程中 的安全性。本文在仿真和实体环境下验证了所提方法的有效性和一定的泛化性, 并完成了引导激光束始终指向运动物体的跟踪实验。

3. 基于CoppeliaSim 仿真与实体机器人的运动目标跟踪和操作平台搭建 本文在CoppeliaSim仿真环境中进行了机器人跟踪并操作运动目标的实验验 证。同时,在真实环境中搭建机器人实验平台,由UR5执行目标跟踪与操作任 务,使用UR3带动目标运动。通过跟踪不同的目标轨迹,验证了所提方法的有 效性并展示了实际的跟踪操作效果。

Other Abstract

Vision is an important way for robots to obtain information from the outside world, and robot motion control based on visual feedback occupies a pivotal position in robotics. As the field of robotics continues to expand, there are more and more scenarios for ma nipulating moving targets, such as grasping moving objects and assembling moving peg-in-hole. The use of vision-guided robots to track and manipulate moving targets has become a hot research topic.

In the vision control techniques for robot manipulation of static targets, the time de lays caused by robot command transmission and image processing are usually ignored. However, in scenarios of robots operating moving targets, the time delay cannot be ig nored, which will degrade tracking performance and even lead to operational failure. In addition, unknown or uncertain system parameters in the robot system can also af fect tracking performance. Therefore, aiming at the problem of model uncertainty, time delay, and other interference factors in the visual control tracking system, the robot tracking controller parameters are optimized based on the current and historical status of the robot tracking system to achieve high-precision moving target tracking and re liable manipulation of the target. The main work and contributions of this paper are summarized as follows:

1. A visual feedback adaptive control system based on delay compensation is proposed. Animage servo-based robot vision tracking method is proposed for ”eye-in-hand” robot tracking system and motion peg-in-hole assembly tasks. To solve the problem of robot tracking accuracy degradation due to uncertain visual servo parameters, a pseudo inverse image Jacobian matrix adaptive algorithm is proposed to adjust the image Ja cobian matrix online. In addition, a time delay compensator based on a radial basis neural network is constructed to eliminate the effects of command transmission and image processing delays in closed-loop tracking systems, thereby reducing tracking er rors caused by time delay. The visual feedback adaptive control system based on delay compensation includes two parts: pseudo inverse image Jacobian matrix adaptive and delay compensation. The effectiveness of the proposed method is verified in simulations and physical experiments, respectively, and the task of moving peg-in-hole assembly is completed when the clearance is only 1 cm.

2. Avisual feedback control system based on reinforcement learning compen sation is proposed. A vision tracking controller based on a sliding mode controller and SiamBAN network is designed for the tracking problem of arbitrarily moving objects in threedimensional space. A servo control perturbation compensation model based on the Actor-Critic reinforcement learning framework is constructed, so as to compensate for tracking errors caused by external perturbations and model errors that cannot be elimi nated by conventional servo controllers. Two kinds of compensators, control input com pensation and reference compensation, are proposed to enhance the visual servo perfor mance based on the sliding mode controller, respectively. The proposed approach com bines the advantages of traditional robot control and reinforcement learning, improves tracking performance through self-adjustment in an online optimization approach and ensures the safety of the robot during operation. The effectiveness and generalization of the proposed method are verified in the simulation and physical environment. The motion tracking experiment is completed, guiding the laser beam to always point at the moving object.

3. An experimental platform for tracking and manipulating motion targets based on CoppeliaSim simulation and physical robots is built. Asimulation environment of CoppeliaSim is built, in which the experimental veri f ication of the robot tracking and manipulating the moving target is carried out. Further, a real-world robot experiment platform is set up in which UR5 performs target tracking and manipulation tasks, and UR3 is used to drive the target into motion. By tracking different target trajectories, the effectiveness of the proposed methods is verified and the actual tracking and manipulation are demonstrated.

Keyword视觉控制,运动目标操作,延时补偿,强化学习
Language中文
Sub direction classification机器人感知与决策
planning direction of the national heavy laboratory实体人工智能系统决策-控制
Paper associated data
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/51865
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
王丽丽. 面向运动目标跟踪的机械臂补偿学习控制技术研究[D],2023.
Files in This Item:
File Name/Size DocType Version Access License
硕士学位论文_王丽丽0531.pdf(9953KB)学位论文 限制开放CC BY-NC-SA
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[王丽丽]'s Articles
Baidu academic
Similar articles in Baidu academic
[王丽丽]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[王丽丽]'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.