毕业生知识库

柔顺控制是机器人领域的重点研究方向之一，在装配和金属加工等需要与对象/环境接触的场景中，以及人机交互作业中具有重要应用价值。阻抗控制通过调整机器人的位置误差与接触力之间的关系，使机器人表现出柔顺特性，保障了作业的舒适性和安全性，是柔顺控制的重要方式之一。

然而，随着智能制造对于机器人在复杂非线性环境中的作业能力提出了更高要求，由于对象/环境的非结构化特性，以及机器人系统的动力学非线性特性，传统的定刚度阻抗控制不能很好地适应复杂的作业任务。因而，如何让机器人能够在人机交互环境、非结构化环境中精确、灵巧地完成作业任务，是目前机器人柔顺控制方向面临的重要挑战。

本文针对环境刚度不确定及目标位置动态变化下的机器人精确、灵巧作业需求，研究非结构环境下的变阻抗学习控制策略，根据环境刚度变化情况在线自适应调节阻抗模型参数，提高了机器人对人机交互环境和非结构化环境的适应能力。本论文的主要工作和创新点归纳如下：
1. 提出面向非结构化零部件精密装配的技能学习与自适应阻抗控制方法
针对非结构化对象的形状不规则、刚度有差异等对机器人造成了非线性扰动的装配问题，本文提出了基于混合动态基元和阻抗参数在线优化的机器人柔顺控制方法。通过混合动态基元实现了机器人的技能学习与泛化，通过机器人动力学模型的线性化，实现了阻抗参数的在线优化。在针对高刚度、低刚度的不规则零件装配中达到了0.03mm的性能。
2. 提出面向人-机力协作装配的变阻抗学习控制模型
针对实际装配条件与理想装配条件存在差异，人-机力协作过程中操作人员技能存在差异的问题，提出利用无模型强化学习方法训练变阻抗学习控制模型，通过人在机器人控制回路的方式在线调整机器人装配动作，将人的感知决策能力和机器人的动作执行能力有效结合，解决针对不同形状物体和高精度装配的问题。
3. 基于CoppeliaSim的装配过程仿真和基于UR3机器人的装配作业平台搭建
本文在搭建的CoppeliaSim装配环境中对变阻抗模型进行训练和仿真验证，并在真实环境中进一步搭建了基于UR3机器人的装配平台，通过多种不同形状物体的实物实验验证所提出方法的可行性和实际装配效果，实验结果表明了方法的有效性。

Compliance control is one of the key research directions in the field of robotics, and has important application value in scenarios that require contact with objects/environments, such as assembly and metal processing, as well as in human-robot interaction. By adjusting the relationship between the position error and the contact force of the robot, impedance control makes the robot show compliance characteristics, which ensures the comfort and safety of the operation, and is one of the important methods of compliance control.

However, as intelligent manufacturing puts forward higher requirements for robots to operate in complex nonlinear environments, due to the unstructured characteristics of objects/environments and the dynamic nonlinear characteristics of robot systems, the traditional constant stiffness impedance control can not well adapt to complex working tasks. Therefore, it is an important challenge for robot compliance control to enable robot to complete the task accurately and flexibly in the human-robot interaction environments and unstructured environments.

Aiming at the precise and dexterous operation requirements of the robot under the uncertainty of environmental stiffness and the dynamic change of target position, this paper studies the variable impedance learning control strategy in the unstructured environments, adaptively adjusts the impedance model parameters online according to the change of environmental stiffness, and improves the adaptability of the robot to the human-robot interaction environments and unstructured environments. The main work and innovations of this paper are summarized as follows:
1. A skill learning and adaptive impedance control method for precision assembly of unstructured parts is proposed
Aiming at the assembly problem that the irregular shape and stiffness difference of unstructured objects cause nonlinear disturbance to the robot, a robot compliance control method based on hybrid dynamic primitives and online optimization of impedance parameters is proposed in this paper. The skill learning and generalization of the robot are realized through the hybrid dynamic primitives, and the online optimization of impedance parameters is realized through the linearization of the robot dynamic model. The performance of 0.03mm is achieved in the assembly of irregular parts with high and low stiffness.
2. A variable impedance learning control model for human-robot collaborative assembly is proposed
In view of the differences between the actual assembly conditions and the ideal assembly conditions, and the differences in the skills of operators in the process of human-robot cooperation, a model-free reinforcement learning method is proposed to train the variable impedance learning control model, and the robot assembly action is adjusted online through human-in-the-loop method, which effectively combines the human perceptive decision-making ability with the robot action execution ability to solve the problems of objects of different shapes and high-precision assembly.
3. Assembly process simulation based on CoppeliaSim and assembly platform construction based on UR3 robot
In this paper, the variable impedance model is trained and simulated in the CoppeliaSim environment, and the assembly platform based on UR3 robot is further built in the real environment. The feasibility and actual assembly effect of the proposed method are verified by physical experiments of various objects with different shapes. The experimental results show the effectiveness of the method.