CASIA OpenIR  > 毕业生  > 博士学位论文
面向多种约束的工业机械臂避障运动规划方法研究
糜凯
2020-05-28
页数166
学位类型博士
中文摘要

      随着“中国智造2025”战略的推进,以及人口老龄化、用工荒问题的逐渐突出,以机器代替人工操作,提高制造业柔性生产能力,推动传统制造业向智能制造转型的需求日益迫切。工业机械臂作为制造业转型升级的核心部件之一,提高其编程效率是加快生产线更新迭代速度、实现柔性化生产的关键。相比于传统的示教编程,在离线编程系统中融合机械臂自主避障的智能规划算法可以显著提高其编程效率。但是由于机械臂自身的非线性和高自由度特性,加上复杂环境下的各种任务约束要求,使得机械臂避障运动规划在工业应用中面临诸多难题。因此,研究多种约束下机械臂避障运动规划方法对提高制造业柔性生产能力、实现智能制造具有重要意义。
      论文结合“机器人关键技术研究及产业化应用”专项(No.161100210300),围绕机械臂运动学建模、碰撞检测与距离计算、避障运动规划等关键问题进行了深入研究,提出了多种具有优化性质的机械臂避障运动规划方法。同时融合环境感知、碰撞检测等功能,开发出了具有自主知识产权的6R型工业机械臂控制系统。本课题完成的主要工作如下:
(1)针对狭窄空间约束下避障运动规划存在轨迹冗余度高、耗时长的问题,提出了基于停滞检测和分享多启发A*搜索的避障运动规划方法。该方法首先在构型空间设计多种运动基元,确保了搜索完备;其次,在环境空间构建启发式引导,提高了引导效率;然后,为了避免不必要的辅助引导增加搜索负担,以关节路径长度为优化目标,设计了基于停滞检测的分享多启发A*搜索机制;最后,为了实现完整的运动规划,研究了基于回溯机制和三次多项式插值的路径后处理方法。仿真及实验结果证明了该方法的有效性,同时针对狭窄空间这一特殊场景,在规划效率和缩减关节路径长度上有着明显优势。
(2)针对末端路径约束下避障运动规划存在末端跟踪精度低、轨迹冗余度高的问题,提出了基于冗余机械臂逆运动学的随机采样优化算法。算法首先采用梯度投影法进行节点扩展,通过在零空间的随机采样,确保了概率完备;其次,采用梯度约减法进行关节轨迹生成,引入负反馈机制,确保满足末端路径约束的同时,实现了关节轨迹的高阶平滑;最后,以关节路径长度和末端跟踪精度为优化目标,在扩展随机搜索树的过程中引入优化结构,提高轨迹质量。仿真及实验结果表明,该算法可以直接规划出一条满足末端路径约束且高阶平滑的避障轨迹,同时有效降低轨迹在关节空间的冗余度。
(3)针对非规则目标域约束下已有算法存在目标位姿随机给定、规划轨迹质量差的问题,提出了基于高斯过程轨迹描述的目标优化算法。算法首先对非规则目标位置域构建符号距离场,推导了相应的距离计算方法及梯度形式;其次,针对目标姿态约束,利用李群李代数进行姿态误差描述,推导了具有加法性质的局部线性扰动模型;在此基础上,定义目标域约束似然,并结合障碍物约束似然和高斯先验分布,将规划问题转化为最大后验概率问题;最后,通过数值优化的方式快速规划出一条平滑的避障轨迹。仿真结果表明,该算法可以将机械臂末端优化至目标域内一个最优或次优位姿处,有效提高了规划轨迹质量。
(4)针对动态目标约束下避障运动规划存在重规划轨迹衔接性差、重规划效率低的问题,提出了基于因子图的分节拍在线运动重规划方法。该方法首先研究了针对目标位姿约束的轨迹优化算法,并将其描述成因子图结构;随后对于未知的动态目标,研究基于iSam2图优化器的分节拍在线运动重规划算法,考虑实际目标感知及轨迹重规划所需要的时间,引入分节拍机制,加快重规划速度,提高了对动态目标跟踪的成功率。仿真结果表明,该方法可以在实现环境避障和动态目标跟踪的同时,快速进行运动重规划,并实现了多条重规划轨迹的平滑衔接。
(5)针对工业机械臂自主避障运动规划要求,开发了6R型工业机械臂控制系统,包括运动控制器、手持示教器和基于ROS的离线编程系统。通过在离线编程系统中融入环境感知、碰撞检测等功能,并结合以上研究的规划算法,实现了给定任务约束下机械臂的自主避障运动规划。实验结果验证了本文所提出算法以及整套控制系统的实用性。

英文摘要

      With the promotion of the strategy "Made in China 2025", as well as the issues of population aging and labor shortage becoming increasingly prominent, the needs to replace manual operations with machines, increase the flexible production capacity of the manufacturing and promote the transformation of traditional manufacturing to smart manufacturing have become increasingly urgent. The industrial manipulator is one of the core components of manufacturing transformation and upgrading, so improving its programming efficiency is the key to speeding up the iteration of the production line and realizing flexible production. Compared with the traditional teaching programming, the offline programming system incorporating intelligent planning algorithms of autonomous obstacle avoidance for manipulators can significantly improve its programming efficiency. However, due to the non-linear and high-degree-of-freedom characteristics of the manipulator and various task constraints in complex environments, obstacle avoidance motion planning of the manipulator faces many problems in industrial applications. Therefore, studying obstacle avoidance motion planning methods of manipulators under various constraints is of great significance to improve the flexible production capacity of manufacturing and realize intelligent manufacturing.
      The thesis is supported by the key technology research and industrialization application of robots under Grant No.161100210300, and researches deeply on the key issues such as kinematics modeling, collision detection and distance calculation, and obstacle avoidance motion planning of manipulators, and proposes multiple obstacle avoidance motion planning methods for manipulators with optimization properties. At the same time, the functions of environment perception and collision detection are integrated to develop a 6R industrial manipulator control system with independent intellectual property rights. The main work completed in this topic is as follows:
(1) Considering the problems of high trajectory redundancy and long time consuming for obstacle avoidance motion planning with narrow space constraints, an obstacle avoidance motion planning method based on stagnation detection and shared multi-heuristic A* search is proposed. First, the method designs multiple motion primitives in the configuration space to ensure a complete search; second, construct heuristic guidance in the environment space to improve the guidance efficiency; then, in order to avoid the unnecessary auxiliary guidance to increase the search burden, a shared multi-heuristic A* search mechanism based on stagnation detection is designed, which takes the joint path length as the optimization goal; finally, in order to achieve complete motion planning, a path post-processing method based on backtracking mechanism and cubic polynomial interpolation is studied. Simulation and experimental prove the effectiveness of the method. At the same time, for the special scenario of narrow space, it has obvious advantages in planning efficiency and reducing joint path length.
(2) Considering the problems of low end tracking accuracy and high trajectory redundancy for obstacle avoidance motion planning with end path constraints, a random sampling optimization algorithm based on inverse kinematics of redundant manipulators is proposed. First, the algorithm uses the gradient projection method to expand the nodes, and ensures the probability is complete by random sampling in zero space; second, it uses the gradient reduction method to generate joint trajectories, and introduces a negative feedback mechanism to ensure that the end path constraint is met while achieving high-order smoothing of the joint trajectories; finally, with the joint path length and end tracking accuracy as the optimization goals, an optimized structure is introduced in the process of expanding the random search tree to improve the trajectory quality. Simulation and experimental results show that the algorithm can directly plan a high-order smooth obstacle avoidance trajectory that satisfies the end path constraints, and effectively reduces the trajectory redundancy in joint space.
(3) Considering the problems of randomly giving goal pose and poor planning trajectory quality for the existing algorithms with irregular goal region constraints, an objective optimization algorithm based on Gaussian process trajectory description is proposed. First, the algorithm constructs a symbolic distance field for the irregular goal position region, and derives the corresponding distance calculation method and gradient form; second, for the goal posture constraint, the Lie Group and Lie algebras is used to describe the posture error, and a local linear perturbation model with additive properties is derived; on this basis, the goal region constrained likelihood is defined, and combined with the obstacle constrained likelihood and the Gaussian prior distribution, the planning problem is transformed into a maximum posterior probability problem; finally, numerical optimization is used to quickly plan a smooth obstacle avoidance trajectory. Simulation results show that the algorithm can optimize the end effector to an optimal or suboptimal pose in the goal region, and improve the planned trajectory quality.
(4) Considering the problems of poor replanning trajectory connectivity and low replanning efficiency for obstacle avoidance motion planning under dynamic goal constraints, a beat-time online motion replanning method based on factor graph is proposed. First, the method studies the trajectory optimization algorithm for the goal pose constraints, and describes it as a factor graph structure; then for the unknown dynamic goals, a beat-time online motion replanning algorithm based on iSam2 graph optimizer is studied. Considering the time required for actual goal perception and trajectory replanning, the algorithm introduces a beat-time mechanism and accelerates the replanning speed to improve the success rate of dynamic goal tracking. The simulation results show that this method can realize the obstacle avoidance and dynamic goal tracking, at the same time, perform motion re-planning quickly, and realize the smooth connection of multiple replanning trajectories.
(5) Considering the requirements of autonomous obstacle avoidance motion planning for industrial manipulators, a 6R industrial manipulator control system is developed, including a motion controller, a handheld teach pendant, and an offline programming system based on ROS. By integrating functions such as environmental perception and collision detection into the offline programming system, and combining the planning algorithms studied above, the autonomous obstacle avoidance motion planning of manipulators under given task constraints is realized. The experimental results verify the practicability of the proposed algorithms and the entire control system.

关键词工业机械臂 避障运动规划 运动学模型 碰撞检测与距离计算 高斯过程轨迹描述
语种中文
七大方向——子方向分类智能机器人
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/39203
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
糜凯. 面向多种约束的工业机械臂避障运动规划方法研究[D]. 中国科学院自动化研究所. 中国科学院大学,2020.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
面向多种约束的工业机械臂避障运动规划方法(20180KB)学位论文 限制开放CC BY-NC-SA
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[糜凯]的文章
百度学术
百度学术中相似的文章
[糜凯]的文章
必应学术
必应学术中相似的文章
[糜凯]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。