【作者】 刘馨；

【导师】 张雪波；

【作者基本信息】 南开大学 ， 人工智能， 2024， 硕士

【摘要】 随着工业自动化和智能化的深入发展,智能自主作业机器人受到了广泛关注,并在智能制造等领域展现出巨大的应用前景。然而,未知复杂环境下机器人在环境感知、执行操作等方面的安全性、自主性、和灵活性仍较低。在此背景下,本文聚焦于机械臂这一关键研究对象,基于“眼在手上”视觉感知系统,围绕用户引导的安全建图、自主探索、实时避障运动规划三个方面开展研究。本文主要工作如下:1.针对现有机械臂半自主建图方法的安全性与执行效率较低的问题,提出一种用户引导的“眼在手上”机器人安全建图与路径规划方法。首先,考虑机器人未观测区域中的潜在碰撞风险,提出一种基于区域划分与占据概率预置的八叉树地图初始化方法,在缩小未知体素搜索范围的同时实现其在地图中的显式表征。而后,通过将未知空间纳入环境碰撞模型中实现机器人的安全建图,同时也为下阶段机器人自主探索提供安全保障。最后,为了快速响应用户下达的目标观测位姿指令,采用一种基于剪枝优化策略的改进RRT-Connect规划算法。实验结果表明,与传统RRT-Connect算法相比,该方法能够在保证运算效率的同时,显著缩短机器人在前往目标位姿时的关节空间移动距离(平均缩短55.2%)。2.针对现有机器人自主探索算法在复杂受限空间中探索效率低的问题,提出一种考虑机械臂构型可操作度的分层式自主探索框架。上层探索框架中,结合了区域边界密度与子地图平均信息熵,以及一种基于光轴余弦距离的新型运动代价,评估各子区域的探索价值;下层探索框架中,创新性地采用了球面斐波那契网格生成均匀覆盖最佳子区域的观测视点。除了视点的信息增益之外,还考虑了与之对应机器人构型的可操作度指标,通过综合评估得到最佳相机-机器人探索构型。实验结果表明,相比于基于视点采样的方法,所提算法有效提高了多自由度机器人自主探索效率(探索时间缩短47.1%,总体移动距离缩短56.9%)。3.针对现有机器人运动规划算法难以兼顾规划前瞻性与反应灵活性的问题,提出一种融合改进RRT-Connect与二次规划的全局-局部实时运动规划方法。首先,通过点云快速欧氏聚类与轴对齐包围盒算法,有效降低环境模型中基础几何体数量,进而提高机器人关节凸包与障碍物之间的碰撞检测效率;在此基础上,利用基于剪枝优化的RRT-Connect算法与B样条曲线得到全局路径;而后,通过求解一个包含机器人关节避障约束与关节速度约束的二次规划问题实现反应式局部运动规划。实验结果表明,所提算法能够成功实现机器人运动过程中考虑避障的在线目标切换并安全抵达新目标。更多还原

【Abstract】 As industrial automation and intelligence deepen,intelligent autonomous operat-ing robots have garnered widespread attention and demonstrated significant application potential in fields such as intelligent manufacturing.However,in unknown and complex environments,the safety,autonomy,and flexibility of robots in aspects such as environ-ment perception and task executions are still relatively low.In this context,this thesis focuses on robotic manipulators,a key research subject,and constructs an“eye in hand”visual perception system.It revolves around three aspects:safety mapping guided by users,autonomous exploration,and real-time obstacle avoidance motion planning.The main contributions of this thesis are as follows:1.Aiming at the low safety and execution efficiency of existing semi-automatic mapping methods for robotic manipulators,a user-guided“eye in hand”manipulator safety mapping method.First,considering the potential collision risks in the unob-served area of the robot,an octomap initialization strategy based on region partitioning and occupancy probability preset is proposed,which reduces the search range of un-known voxels while achieving their explicit representations.Consequently,the safety mapping of the robot is addressed by incorporating them directly into the environment collision model.Finally,to rapidly respond to the target observation pose instructions given by users,an improved RRT-Connect algorithm with a pruning strategy is pro-posed.The experiment results show that compared with the traditional RRT-Connect algorithm,the proposed path planning method can significantly shorten the joint space movement distance of the manipulator(an average reduction of 55.2%)while ensuring computational efficiency.2.Aiming at the low efficiency of existing autonomous exploration algorithms for manipulators in complex and confined spaces,a hierarchical autonomous exploration framework considering the manipulability of robot configurations is proposed.In the upper layer of the framework,the exploration potential of each subregion is evaluated by combining the density of frontiers with the average information entropy of all vox-els inside,as well as a novel motion cost based on the cosine distance of the optical axis;In the lower layer,a spherical Fibonacci lattice is adopted to generate uniformly distributed viewpoints covering the best subregion,and the best exploration configura-tion is determined by comprehensively assessing the information gain of the viewpoints and the manipulability indices of the corresponding robot configurations.The exper-iment results show that compared with the SOTA method,the proposed algorithm ef-fectively improves the autonomous exploration efficiency of high Degree-of-Freedom(Do F)robots(reducing exploration time by 47.1%and overall movement distance by56.9%)3.Aiming at the problem that existing robot motion planning algorithms are diffi-cult to balance planning foresight and reaction flexibility,a global-local real-time mo-tion planning method combining improved RRT-Connect and quadratic programming is proposed.First,by adopting the fast Euclidean clustering and axis-aligned bounding box algorithm,the number of geometric bodies in the environment model is signifi-cantly reduced,thereby the collision detection efficiency between robot joint convex hulls and obstacles is improved;On this basis,the improved RRT-Connect algorithm based on pruning strategy and B-spline curve are used to obtain the global path;Then,reactive local motion planning is achieved by solving a quadratic programming problem that includes robot joint obstacle avoidance constraints and joint velocity constraints.The experiment results show that the proposed algorithm can successfully achieve robot online target switching and safely reach the new target with obstacle avoidance taken into consideration.更多还原