Research On Motion Planning Method Of Autonomous Underwater Vehicles Based On Multiobjective Constraint

Autonomous Underwater Vehicles(AUVs)have become an important tool for underwater missions due to their characteristics of small size,wide operation ranges and equipped with sensors with different functions.AUV with autonomous motion planning ability can complete various missions through behavior planning and motion planning,widely used in underwater patrol and reconnaissance,target detection and recognition,hydrology and water quality information detection and other aspects.As the Marine environment is a complex dynamic system,AUV is confronted with complex environments and situations such as sudden obstacles,unknown water environment,counter-reconnaissance and counter-rounding when performing.In addition,the underwater environment has great restrictions on signal transmission,and AUV needs to rely on its own management to complete the mission entrusted by human beings,which is inseparable from its motion planning ability.Based on the background of AUV regional scanning in broad water area and mapping in river water area,in order to improve the AUV’s underwater efficiency and autonomous mission capability,how to improve the effect of exercise planning is considered as the main problem,and the following contents are studied:(1)On the basis of summarizing the previous research on the structure of mobile vehicles,considering the operation characteristics of AUV and based on the idea of hierarchical reinforcement learning hierarchy,a dual structure of "task decomposition-behavior planningaction planning" and "environment interaction-global planning-local planning" is established for AUV motion planning.Establish hierarchical principles for top-down decision-making and bottom-up learning.Set the tone for planning strategies through learning and training.A complete planning process for AUV underwater operation is presented.(2)Considering the characteristics of ocean environment,AUV and mission requirements,a multi-objective constraint model is established,which took economy and safety as the main objectives,and the dynamics and kinematics of AUV and ocean environment as the main constraints.The mathematical models of constrained optimization problems and multiobjective optimization problem are analyzed.The decoupling of constraints is taken as the entry point of AUV multi-objective constraint problem,and the constraint model is simplified.The problem of mutual restriction of various objectives in AUV task is solved.(3)The task decomposition and behavior planning of AUV search in broad water area under multi-objective constraint are studied,that is,to achieve the global planning of behavior and complete the first two layers of the structure.The search task in broad water area is decomposed into principal axis optimization behavior,search area traversal behavior and whole area scanning behavior.The three behaviors are constrained and optimized respectively.Combined with the scope of AUV operating environment,kinematics constraints,environmental constraints,etc.,the energy consumption and safety in the process of AUV operation are comprehensively considered,and the optimal search direction is obtained.Compare depth-first search and breadth-first search strategies to determine the search area traversal scheme.The deep reinforcement learning algorithm is improved based on the prioritized experience replay,which solves the problem of slow convergence.The advantages of the algorithm are verified through comparative training.The comb path scanning strategy of AUV system is trained by using target constraint as reward function of reinforcement learning.The feasibility and stability of the proposed algorithm are verified by simulation experiments.At the same time,the effectiveness of the proposed path was verified in the field experiment of searching a wide water area.(4)The task decomposition and behavior planning of river waters mapping by AUV under multi-objective constraints are studied.The mapping task of river waters is decomposed into path optimization behavior and travelling salesman optimization behavior,and then the two behaviors are constrained and optimized respectively.Genetic algorithm is improved based on node optimization strategy.Genetic operators are designed based on the constraints in AUV operation.The optimal path between cross-sectional nodes of river is obtained through algorithm iteration,which solves the problems of energy consumption increase and AUV stranding easily caused by too many path nodes planned by traditional algorithm.According to the actual operation requirements,the travel salesman problem model is established.Based on the continuous Hopfield neural network algorithm,the traversal optimization of all crosssectional nodes is realized,and the problem of missing measurement points in surveying and mapping process is avoided.In this paper,different river models are established and global behavior planning under multi-objective constraints is carried out according to their different characteristics.The optimal navigation path of AUV in the mission is obtained.The feasibility of planning path is verified by field test.(5)The action planning and action execution of AUV under multi-objective constraints are studied based on multi-objective method,that is,local action planning is carried out to complete the last layer of the structure,realize the tracking of the global path and the avoidance of unexpected obstacles.The Deep Deterministic Policy Gradient(DDPG)algorithm was improved based on the artificial experience pool strategy,and the action planning strategy of AUV was trained,and the "state-action" mapping relationship was obtained.Based on this mapping relationship,the system can output action instructions in the corresponding state to guide the AUV towards the target or avoid obstacles.Through the comparison test,the proposed algorithm is verified to be faster convergence and more stable in the later training period,which solves the problems of mechanical rigidity of AUV action and large planning error in the field test by using the conventional algorithm.At the same time,the improved reinforcement learning algorithm based on critic neural network trains the "action-execution" mapping relation,and solves the problem of low learning efficiency of the algorithm with the 6-dof motion characteristics of AUV.The policy is used to control the AUV to achieve the target action obtained in the previous step.The "state-motion-execution" decision-making process is obtained by integrating the above two learning and training steps,which solves the problem of high energy consumption caused by AUV snaking tracking and large angle obstacle avoidance brought by the "end-to-end" strategy.Based on the designed strategy,the simulation experiment of AUV motion planning is carried out to verify the effectiveness and stability of the algorithm,which reflects the advantages of the algorithm.Under the background of the whole area search task and river mapping task,the process of planning structure is completed.