Research On Key Technology Of AUV Path Planning Issues

The genuine meaning of autonomy capability of AUV (Autonomous Underwater Vehicle) is to process the ability to interact with external environment, one major respect of this interaction is for AUV to be capable of global path planning, dynamic re-planning when encountering abrupt events, and obstacles avoiding. For AUV to fulfill autonomous voyage and task execution, path planning is an important sector, therefore, research on path planning technology of AUV plays an important role in improving its intelligent level and accelerating its engineering application. With the support of the key research project (4131607) of the 10th five-year plan and project (51316080202) of the 11th five-year plan, the thesis deeply studied AUV's global path planning, local path planning, obstacle avoidance planning and path planning for region reconnaissance in the ocean.Firstly, the kinetics model and system components of AUV are introduced, and the current research situation about AUV architecture is summarized. The hierarchical control architecture is designed for certain AUV, according to the laws of increasing precision with decreasing intelligence (IPDI). The architecture consists of six layers and three lists, and the autonomous control realized by the architecture can satisfy the needs of AUV task.Secondly, we bring forward the methods for AUV global path planning. We devises shortest tangent path algorithm used for small region path planning that computes simply and runs efficiently. As to large scope ocean area, we design a multi-objective optimization genetic algorithm based on VCF electronic charts. This algorithm encodes chromosome into real-number coordinates of variable length, and takes several influence factors into account when designing evaluation functions. Infeasible path is given an additional punishment and then participates in population evolution. Five genetic operators such as selection, crossover, mutation, repair and deletion operators are designed. Domain knowledge is introduced in the process of initial population generation and genetic operator design, in order to achieve the goal that the generated paths do not cross over obstacle area to the greatest extent. Convergence speed of this algorithm is greatly improved as a result. So an approximate optimal feasible path in large area can be generated as quickly as possible using this algorithm.Thirdly, we put forward local path planning methods based on fuzzy logic. Considering the huge indeterminacy and inaccuracy of data from navigation system and sensors under complex ocean condition, this method can be used to deal with the approximate information. After analysis of the relationship among ocean current, obstacles and AUV, the derived fuzzy rules tables are used as basis of local path planning. This method considers both ocean current effect and motion characteristics of AUV platform, thus it can well solve the local path planning problem considering ocean current effect in a relatively simple form. Local minimum problem also arises in the application of fuzzy control theory to AUV path planning. So a real-time path planning method based on sensor information is proposed, and it enables AUV to escape from local minimum state and consequently arrive at target position by setting up virtual target point together with activation and exit condition of follow-left-wall behavior and follow-right-wall behavior.Fourthly, we present a method for automatic design of reactive behavior based on sensor information, which regards collision avoidance planning as an integrated behavior without the need to consider how to decompose behaviors. Q learning is utilized in the independent learning about AUV reactive behavior so as to obtain the anticipant optimal behavior. The control rules extracted after learning process completion can be applied to direct reactive behaviors. In addition, a fuzzy Q(λ) algorithm is presented on system's request of continuous input and output space. It combines fuzzy theory with Q leaning, utilizes Q(λ) algorithm to update result vector weights continuously until they are converged, thereby, a complete fuzzy rule base is produced.Last, we put forward complete coverage path planning algorithms. Weight gradient method and cell decomposition method are devised for basic region. As for common region that contains obstacles, the whole region is divided into several sub-regions, and cell decomposition strategy based on critical points is presented. An IN/OUT critical points matching algorithm is proposed to tackle the problems arisen during adjacency graph construction, which guarantees the entire search for all sub-regions. In order to follow the planned path, a path following strategy for AUV is designed which considers both task requirements and AUV kinematic ability to acquire objective instructions on AUV movement. Experiment results of AUV execution of region reconnaissance task verify the proposed algorithm which meets the requirements of practical application.