基于Kinect的移动机械臂平台开门策略设计与实现

目前对于机器人开门方法的研究基本都是在机械臂位于门前固定位置的前提下,通过对门把手进行识别来实现的.将机械臂与机器人本体运动控制相结合,提出了一种基于深度传感器(Kincet)的移动机械臂开门策略.给出开门策略的流程图,分析基于Kinect的图像接收处理过程,采用色调饱和度强度(HSI)颜色空间对门把手进行定位的方法以及将门把手定位信息与五自由度机械臂的运动控制相结合的控制策略.在VS2008环境下编写控制软件,以开门实验验证了该方法的有效性.     

基于深度强化学习的水下机械臂自主控制研究

受限于水下环境和观察角度,精确地控制水下机械臂完成水下自主作业任务具有很大的挑战性。针对这一问题,考虑到强化学习算法对环境具有强大的自适应能力,文章提出一种水下机械臂自主控制方法。首先,其设计了一种近端策略优化（PPO）结合行动器-评判器（AC）的强化学习算法训练自主控制策略;其次,对训练过程中出现的稀疏奖励问题,提出了一种基于人工势场法的奖励塑造方法;最后,在仿真试验环境中对基于上述方法训练得到的自主控制策略进行了验证。验证结果表明,该策略可以有效收敛并自主控制水下机械臂向目标位置精确移动,移动快速、过程流畅,末端执行器运动轨迹平滑、稳定。     

基于三步法的机械臂轨迹跟踪控制

考虑机械臂末端轨迹跟踪控制问题,以跟踪逆运动学求解出的末端期望轨迹对应的各关节期望角度为控制目标.设计了一种基于三步法的控制器,该控制器由类稳态控制、可变参考前馈控制和误差反馈控制3部分组成.证明了该控制器可以通过控制机械臂的各关节力矩实现各关节实际角度对期望角度的状态跟踪,进而使得末端轨迹渐近跟踪期望轨迹,并且跟踪误差是输入到状态稳定的.仿真表明基于三步法控制器的空间机械臂末端可以渐近跟踪期望轨迹,并且该算法可以克服系统的末端负载质量变化等不确定性的影响.     

多智能体强化学习的机械臂运动控制决策研究

针对传统运动控算法存在环境适应性较差,效率低的问题。可以利用强化学习在环境中不断去探索试错,并通过奖励函数对神经网络参数进行调节的方法对机械臂的运动进行控制。但是在现实中无法提供机械臂试错的环境,采用Unity引擎平台来构建机械臂的数字孪生仿真环境,设置观察状态变量和设置奖励函数机制,并提出在该模型环境中对PPO(proximal policy optimization)与多智能体（agents）结合的M-PPO算法来加快训练速度,实现通过强化学习算法对机械臂进行智能运动控制,完成机械臂执行末端有效避障快速到达目标物体位置,并通过该算法与M-SAC（多智能体与Soft Actor-Critic结合）和PPO算法的实验结果进行分析,验证M-PPO算法在不同环境下机械臂运动控制决策调试上的有效性与先进性。实现孪生体自主规划决策,反向控制物理体同步运动的目的。     

协作机械臂碰撞环境下的安全控制

针对人机协作过程中机械臂末端与操作者之间可能存在物理碰撞的问题,本文提出一种基于自适应导纳和滑模控制的机械臂安全控制方法.首先,采用基于广义动量的干扰观测器估计机械臂各关节和末端执行器上的干扰力.然后,设计一个双环控制结构,内环是基于干扰观测的双幂次趋近律滑模轨迹跟踪控制器,用以克服外界干扰的影响,达到期望的轨迹跟踪精度;外环是自适应导纳控制器,依据碰撞外力的大小和方向更新期望轨迹,以保证机械臂碰撞发生时的柔顺性和碰撞解除后迅速恢复工作状态的能力.最后,对所提算法进行仿真和实验验证.实验结果表明,协作机械臂碰撞环境下的安全控制算法能保证人机协作的安全性和机械臂工作的连续性,满足人机协作过程中的工作需求.     

基于视/力觉反馈的复合定位方法研究

视觉引导是机械臂的一种常用工作模式,具有结构简单、开发周期短、效率高的特点。然而对于高精度工件的装配而言,需要较高的相机分辨率,从而极大地增加项目预算。针对高精度工件的装配需求,提出了一种基于视/力觉反馈的复合定位方法。该方法分为粗引导和精定位2个阶段：在粗引导阶段,通过视觉系统将工件引导至基座平面的导向范围内;在精定位阶段,利用六维力传感器将工件移动至期望位置,从而完成装配。搭建了基于六轴机械臂的验证系统并进行了验证实验。结果表明,所提出基于视/力觉反馈的复合定位方法适用于高精度工件的装配。     

基于PMSM驱动的柔性关节空间机械臂轨迹跟踪控制方法

针对目前柔性关节空间机械臂轨迹跟踪控制方法忽略了不同重力影响下的机械臂驱动力变化,导致柔性关节空间机械臂轨迹跟踪控制效果较差的问题,提出了基于PMSM驱动的柔性关节空间机械臂轨迹跟踪控制方法。基于构建PMSM驱动数学模型,采用PMSM的矢量控制方法,分析驱动力矩矢量。根据驱动力矩矢量分析结果,分析不同重力环境下有、无摩擦时的驱动力矩。构建柔性关节模型,分析其在不同重力环境下遇到的重力释放问题,使用自适应反演滑膜控制方法,设计控制率,保证机械臂能够按照既定的方向运动,使机械臂具有鲁棒性。根据柔性关节空间机械臂动力学特性,分析不同重力环境下基于PMSM驱动力矩,确定重力项是随之发生改变的。设计控制器,构建动力学模型,确保空间阶段能够最大限度跟踪运动轨迹。实验结果表明,所提方法X轴、Y轴的末端跟踪结果均与实际运动轨迹一致,误差为0。关节控制力矩在时间为3s时,出现了最大为0.5N.m的误差,说明所提方法的跟踪控制效果较好。     

机械臂在未知环境下基于力/位/阻抗的多交互控制

针对人-机械臂交互的不确定性以及环境的未知性,提出了一种基于速度控制的方法来研究人、机械臂和环境三者之间的交互问题.首先通过在机械臂末端安装2个6维力传感器来实现交互力的检测,并在末端工具连接处增加单自由度柔性环节来实现机械臂的柔顺性.其次,基于i Ta SC(instantaneous task specification using constraints)方法对机械臂与工具系统的柔顺度建模,在输出空间中建立起速度/位置与力的关系,实现机械臂与环境之间接触力的控制,通过刚度矩阵实现人对机械臂的拖动控制.另外结合力/位/阻抗混合控制方法实现了在安全区域约束下人拖动机械臂来完成一系列基于位置/力的任务,并与未知环境之间达到一个良好的交互.最后将提出的算法在UR5机械臂上进行轮廓跟踪任务的验证,实验结果表明该方法能达到一个较好的交互控制效果.     

基于虚拟力引导的人机协同目标抓取方法

针对力反馈遥操作中传统人工势场法无法适应于机械臂整体的避障以及在作业过程中操作者难以控制机械臂到达所需位姿的问题,提出了一种基于虚拟力引导的人机协同目标抓取方法。力反馈设备向操作者提供力觉交互。通过结合人工势场法和虚拟夹具,构建管道形虚拟力场,生成实时虚拟力引导,实现协助操作者完成从端机器人的整体避障任务并在完成避障后引导机器人返回预定义路径并趋近目标点。当进行抓取任务时,构建锥形虚拟力场,实现协助操作者操作机械臂到达目标位置和姿态。此外,提出了一种机器人运动限制方法以降低操作者的操作失误对抓取任务的影响。实验证明,该方法能有效提高目标抓取操作的成功率和操作效率。     

面向机械臂轨迹规划的强化学习奖励函数设计

针对基于深度强化学习的机械臂轨迹规划方法学习效率较低,规划策略鲁棒性差的问题,提出了一种基于语音奖励函数的机械臂轨迹规划方法,利用语音定义规划任务的不同状态,并采用马尔科夫链对状态进行建模,为轨迹规划提供全局指导,降低深度强化学习优化的盲目性。提出的方法结合了基于语音的全局信息和基于相对距离的局部信息来设计奖励函数,在每个状态根据相对距离与语音指导的契合程度对机械臂进行奖励或惩罚。实验证明,设计的奖励函数能够有效地提升基于深度强化学习的机械臂轨迹规划的鲁棒性和收敛速度。     

Motion Control of a Nonholonomic Mobile Manipulator in Task Space

In this paper, the motion control of a mobile manipulator subjected to nonholonomic constraints is investigated. The control objective is to design a computed‐torque controller based on the coupled dynamics of the mobile manipulator. The proposed controller achieves the capability of simultaneous tracking of a reference velocity for the mobile base and a reference trajectory for the end‐effector. The aforementioned reference velocity and trajectory are defined in the task space, such task setting imitates the actual working conditions of a mobile manipulator and thus makes the control problem practical. To solve this tracking problem, a steering velocity is firstly designed based on the first‐order kinematic model of the nonholonomic mobile base via dynamic feedback linearization. The main merit of the proposed steering velocity design is that it directly utilizes the reference velocity set in the task space without requiring the knowledge of a reference orientation. A torque controller is subsequently developed based on a proposed Lyapunov function which explicitly considers the coupled dynamics of the mobile manipulator to ensure the mobile base and end‐effector track the reference velocity and trajectory respectively. This proposed computed‐torque controller is able to realize asymptotic stability of both the base velocity tracking error and the end‐effector motion tracking error. Simulations are conducted to demonstrate the effectiveness of the proposed controller.     

Multi-digit maximum voluntary torque production on a circular object

Individual digit-tip forces and moments during torque production on a mechanically fixed circular object were studied. During the experiments, subjects positioned each digit on a 6-dimensional force/moment sensor attached to a circular handle and produced a maximum voluntary torque on the handle. The torque direction and the orientation of the torque axis were varied. From this study, it is concluded that: (1) the maximum torque in the closing (clockwise) direction was larger than in the opening (counter clockwise) direction; (2) the thumb and little finger had the largest and the smallest share of both total normal force and total moment, respectively; (3) the sharing of total moment between individual digits was not affected by the orientation of the torque axis or by the torque direction, while the sharing of total normal force between the individual digit varied with torque direction; (4) the normal force safety margins were largest and smallest in the thumb and little finger, respectively.     

Multi-digit maximum voluntary torque production on a circular object

Individual digit-tip forces and moments during torque production on a mechanically fixed circular object were studied. During the experiments, subjects positioned each digit on a 6-dimensional force/moment sensor attached to a circular handle and produced a maximum voluntary torque on the handle. The torque direction and the orientation of the torque axis were varied. From this study, it is concluded that: (1) the maximum torque in the closing (clockwise) direction was larger than in the opening (counter clockwise) direction; (2) the thumb and little finger had the largest and the smallest share of both total normal force and total moment, respectively; (3) the sharing of total moment between individual digits was not affected by the orientation of the torque axis or by the torque direction, while the sharing of total normal force between the individual digit varied with torque direction; (4) the normal force safety margins were largest and smallest in the thumb and little finger, respectively.     

空间机械臂全局反作用优化及其地面试验研究

针对空间机械臂在运动过程中对其基座产生力／力矩反作用这一问题进行了分析,并提出一种关节轨迹参数化的全局反作用优化方法,用于减小机械臂的反作用力／力矩．基于气浮轴承方法,建立了3自由度空间机械臂反作用优化地面实验系统．最后,采用3自由度机械臂验证了该方法的有效性．     

A framework for large‐force task planning of mobile and redundant manipulators

A framework tackling the problem of large wrench application using robotic systems with limited force or torque actuators is presented. It is shown that such systems can apply a wrench to a limited set of Cartesian locations called force workspace (FW), and its force capabilities are improved by employing base mobility and redundancy. An efficient numerical algorithm based on 2ⁿ‐tree decomposition of Cartesian space is designed to generate FW. Based on the FW generation algorithm, a planning method is presented resulting in proper base positioning relative to large‐force quasistatic tasks. Additionally, the case of tasks requiring application of a wrench along a given path is considered. Task workspace, the set of Cartesian space locations that are feasible starting positions for such tasks, is shown to be a subset of FW. This workspace is used for identifying proper base or task positions guaranteeing task execution along desired paths. Finally, to plan redundant manipulator postures during large‐force‐tasks, a new method based on a min–max optimization scheme is developed. Unlike norm‐based methods, this method guarantees no actuator capabilities are exceeded, and force or torque of the most loaded joint is minimized. Illustrative examples are given demonstrating validity and usefulness of the proposed framework. ©1999 John Wiley & Sons, Inc.     

Task demands and human capabilities in door use

Doors are ubiquitous in the built environment, bur despite their frequent use by people and their involvement in over 300,000 injuries per year (USA), they are little studied from a human factors perspective. This paper provides a classification scheme for doors based on human/door interaction and a prototypical task analysis of door use. Two observational studies were performed on a particular type of door. The first observed 1600 human/door interactions and found that people's use of force-enhancing strategies increased for larger doors, particularly for people of smaller stature. The second observed 800 interactions with push doors and found that the pont where force is exerted is higher for taller individuals and closer to the center of the door than is typically assumed for placing handles. Th second study was partly corroborated by measuring the position of wear patterns on doors. As is expected in human factors, the was people used doors was a function of both task demands and human capabilities. These results suggests that those specifying doors for building s use a restoring torque below 30 Nm and site the handle or push plate 250-350 mm from the door edge and 1000-1500 mm above the floor.     

空间机器人随机故障容错规划的蜜蜂算法

本文为解决复杂的随机规划问题设计了一种基于随机模拟的混沌量子蜜蜂算法,证明了该算法的收敛 性,并分析了算法的收敛速度．分析6 自由度空间机器人系统的不确定性,采用基于微分变换法进行误差分析,建 立了随机数学规划模型．为涉及故障前后运动学与动力学约束限制的容错轨迹规划,以加权最小驱动力矩为优化性 能指标,采用混沌量子蜜蜂算法求解全部工作时间中机械臂故障前后的最优轨迹．通过降低异常关节的运动速度来 降低故障关节力矩,保证机械臂在发生故障后具有较高的操作能力．案例研究验证了该算法的有效性、稳定性及准 确性．