【作者】 徐啟蕾；

【导师】 唐功友；

【作者基本信息】 中国海洋大学 ， 计算机应用技术， 2013， 博士

【摘要】 自主移动智能体（AutonomousMobileAgent）是智能体研究和工程应用的一个重要领域，对于移动智能体而言，其自主导航能力是智能体各个能力中重要的一部分。自主导航包括路径规划、避障策略、定位、环境感知等多个方面，要求移动智能体不具备任何先验的环境知识，仅根据自身配备的传感器来感知周边环境内的障碍物信息，并根据此信息自主规划它的动作、制定运动轨迹，最终尽可能高效、可靠地从初始位置（或初始位姿）到达指定的目标位置（或终止位姿）。移动智能体自主导航需要满足一定的约束条件，这些约束条件包括：与障碍物间无碰撞、路径长度最短、能源消耗最少、时间消耗最少、以及移动智能体的动力学约束等。本文主要研究了移动智能体在大尺度范围内和小尺度范围内未知环境下的自主导航问题，主要的研究内容和创新点现概括如下：1.提出矢量化路径描述方法，以路径矢量组来表示路径。该方法中提出以一组首尾相连的矢量来描述、保存规划路径，移动智能体在根据矢量化的路径行进时，可以直接根据规划路径执行前进、转向指令。而且该方法可以有效的避免路径规划过程对地图构建的依赖，无需地图存储及维护的庞大数据，因此可以大幅度降低保存路径时的空间需求。提出的路径更新计算方法可方便的进行路径的维护和更新，并大大减少路径更新时的计算量。2.将矢量化路径描述方法与Bug算法思想相结合，解决大尺度范围内静态未知环境下移动智能体自主导航问题。研究中借鉴了Bug系列算法中“朝向目标位置移动”和“遇到障碍物转身”的思想，并将其与矢量化路径描述方法紧密结合。首先以起始位置和目标位置构建初始路径，令移动智能体沿初始路径行进，在行进过程中，移动智能体根据不同的条件对未知的环境及其中的障碍物进行检测，一旦发现有障碍物阻挡当前行进路径，则生成中间插入点并更新路径，然后令移动智能体沿更新后的路径继续行进，最终令移动智能体无碰撞的到达目标位置。研究中提出了两种中间插入点的生成方法，分别是扫描生成和随机生成。两种方法中，扫描方式相比于随机方式而言，更容易得到路径长度较优的规划结果；而随机生成方式则具有更好的全局搜索能力。对于扫描生成中间点时易得到局部最优解的缺点，研究中提出了改进方法，使得移动智能体在陷入局部最优后，能够迅速走出局部最优；而对于中间点的随机生成方式则提出了快速收敛策略，以得到较短的运行轨迹和较少的规划时间。3.将矢量化路径描述方法运用到大尺度范围内动态未知环境下移动智能体的自主导航研究中，解决环境中具有未知移动障碍物的自主导航问题。依据矢量化路径描述方法的思想，首先建立初始路径，令移动智能体在沿路径行进的过程中以规定的时间间隔对周边环境进行检测，查找是否有障碍物，一旦发现有障碍物存在，则根据研究中提出的一系列步骤，判定障碍物的移动趋势，并断定该障碍物是否会阻挡当前行进路径。一旦发现有障碍物已经阻挡或将有可能阻挡当前路径时，根据提出的中间点生成算法，朝向障碍物移动的相反方向生成中间插入点并实时更新路径，然后令移动智能体沿更新后的路径行进，直至安全无碰撞的到达目标位置。4.充分考虑移动智能体的动力学约束，提出基于Bug思想的小尺度范围内移动智能体自主导航算法。将Dubins路径与Bug算法相结合，在移动智能体从一个位姿向另一个位姿运动的过程中，根据移动智能体的最小转向半径构建Dubins路径，并令移动智能体沿Dubins路径行进，以满足其动力学特性。首先令移动智能体沿Dubins路径从初始位姿向终止位姿移动，并在规定的条件下感知环境信息，一旦移动过程中发现有障碍物阻挡当前路径，则根据提出的中间位姿生成方法，生成中间位姿，并构造两条分别由当前位姿到中间位姿、由中间位姿到终止位姿的Dubins路径，令移动智能体从当前位姿沿该路径向终止位姿移动。移动过程中不断重复移动、感知、生成中间位姿等一系列过程，直至移动智能体无碰撞的到达终止位姿。更多还原

【Abstract】 Research of the autonomous mobile agent is an important field for the research and engineering application of the agent. For a mobile agent, its ability of autonomous navigation is an important part of its various capacities. Autonomous navigation for the mobile agent includes online path planning, obstacle avoidance, localization of the mobile agent, and environment and obstacle perception, etc. It requires the mobile agent with no prior information of environment or obstacles, only perceiving surrounding environment information according to its own perception equipment, and making the mobile agent autonomously plan its movement and moving trajectory based on its sensing information. Eventually, the mobile agent can autonomously move from the starting position (or initial configuration) to the specified goal position (or final configuration) as efficient and reliable as possible. Autonomous navigation for the mobile agent requires the agent meeting certain constraints, such as collision-free with the obstacle, the shortest path length, the lowest energy consumption, the shortest time cost, and kinematic constraints, etc.This paper studies the autonomous navigation problem for the mobile agent in unknown environment within the scope of large-scale and small-scale. Major researches and innovations are summarized as follows: 1. The concept of vectorization path is proposed in this research, and the planned path is described in the form of vector set. All the planned paths are described and saved in a set of end-to-end vectors. Different from other path representation, when the mobile agent moves along the vectorization path, it can directly execute the moving and steering instructions according to the planning path. And the proposed method can effectively avoid the dependence on map construction in the planning process. There is no huge data requirement for map storage and maintenance, so the space requirement for path saving can be reduced significantly. And the proposed path regeneration method can facilitate the maintenance and updating of the path, and greatly reduce the amount of computation on path updating.2. This research combines the vectorization path representation and the thinking of Bug algorithm in order to solve the autonomous navigation problem for the mobile agent in static unknown environment within the scope of large-scale. The study draws on the move-to-goal and boundary-following thinking in Bug family, and combines them firmly with the vectorization path thinking. Firstly, construct the initial path in accordance with the starting position and the goal position. And then make the mobile agent moving along the initial path. In the process of moving, the mobile agent perceives the obstacles in the completely unknown environment depending on the different conditions. Once there is obstacle constructing the current path, an intermediate point will be generated and inserted into the current path to get a regenerated collision-free path. And then make the mobile agent move along the regenerated path, so as to arrive at the goal position with no collision. In this research, two intermediate point generation methods are proposed, which are scanned generating method and randomly generating method. Compare to random generating method, scanned generating method can get a shorter path length, but random generating method has better global search ability. The shortcoming of scanned generating method is that the mobile agent is easily getting the local optimal solution. The research proposed improving strategy for this shortcoming in order to make the agent quickly out of the local optimum after it falling into the local optimum. For the random generating method, the rapid convergence strategy is proposed to get shorter path trajectory and executing time.3. The vectorization path representation is applied to the research of autonomous navigation for mobile agent in dynamic unknown environment within the scope of large-scale, in order to solve the path planning problem in the environment with moving obstacles. According to the vectorization path representation thinking, establish the initial path firstly, and make the mobile agent moving along the path with perceiving the environment at certain time interval. Once the obstacle is found, the mobile agent will determine the moving trend of the obstacle, and make a judgment of whether the obstacle will construct the current path or not. If there is or will be an obstacle constructing the current path, an intermediate point, in the opposite direction to the moving direction of obstacle, will be generated in accordance with the proposed strategy. And then the collision-free path will be regenerated by inserting the intermediate point. The mobile agent will move along the regenerated path to arrive at the goal position collision-free.4. Taking full consideration of the kinematic models and constraints of the mobile agent, a novel autonomous navigation method for mobile agent in the range of small-scale based on Bug thinking is proposed. Combining the Dubins path with the Bug family algorithm, when the mobile agent moving from one configuration to another, it always moving along the Dubins path which is made up in accordance with the agent’s minimum turning radius, in order to meet the kinematic constraints of the mobile agent. First of all, make the mobile agent moving from the initial configuration to the final configuration along the Dubins path. Perceive the obstacles in environment under certain conditions while moving. If there is obstacle constructed the current path, generate the intermediate configuration in accordance with the proposed strategy, and make the mobile agent move along the new Dubins path from current configuration to intermediate configuration. The Dubins path from intermediate configuration to the final configuration is also generated. Repeat the series of process, such as moving, perception and intermediate configuration generation, until the mobile agent arrives at the final configuration with no collision.更多还原