Research On Task Assignment Method For Multi-AUV System Based On SOM Algorithm

Improving the efficiency of ocean exploration and enhancing the ability to control the ocean is the urgent need to implement the national major strategy of "Marine power".As a strong autonomous mobile underwater platform,underwater robots have become an important tool in accomplishing various complex underwater tasks.The task assignment is the core problem of multi-robot collaboration,also the key to maximize the collaboration capability of multi-robot systems.However,it is difficult for the existing task assignment models to meet the actual task scenario requirements by ignoring environmental constraints such as ocean currents and focusing on a single mission scenario with a single type of robot.Meanwhile,current task assignment methods suffer from slow convergence speed and low solution accuracy.Therefore,the main content of this paper is to propose SOM networkbased task assignment algorithms with high learning efficiency,consistent solution and strong applicability to heterogeneous systems for the existence of large-scale marine operation areas,mixed task scenarios and heterogeneous underwater robot systems.Firstly,an improved SOM algorithm is proposed for the task assignment problem of multi-AUV(Autonomous Underwater Vehicles)systems under the influence of ocean currents in emergency task scenarios.Considering the influence of ocean currents on the work cost,the cruising time is adopted to substitute the cruising distance to measure the work load of each AUV as the basis for achieving work load balance.A neuron population optimization strategy is proposed to improve the learning efficiency and convergence speed of the algorithm by adding neurons with high mapping ability and removing neurons with low mapping ability based on the proposed equation for quantitative measurement of neuron mapping ability.Further,an adaptive workload balance mechanism is proposed to achieve a good load balance among multiple AUVs by considering the maximum work capacity of AUVs and the consumed work cost,thus improving the duration of the AUV system.Then,an RSOM(Reinforced SOM)algorithm is proposed to achieve a consistent solution for the task assignment problem with a mixture of emergency task scenarios and non-emergency task scenarios.The S-TSP(Situation-traveling salesman problem)is defined to integrate the two scenarios into a unified optimization problem by considering the constraints and optimization objectives in both scenarios.The regional learning rate is proposed by considering the individual task value and the task set topology relationship.The response and assignment of tasks in terms of regions rather than single-task preferences achieve the S-TSP defined in a consistent solution.The HIG(Historical information guidance)mechanism is proposed to achieve the improvement of mapping ability and learning efficiency by enriching the update information with historical information in neuron update process.Meanwhile,the proposed workload balance mechanism considers the working capacity and energy consumption of each AUV to improve the continuous working capacity of multi-AUV systems.Simulations at different scales verify the superior performance of the proposed RSOM algorithm for consistent solution of mixed task scenarios.Lastly,a task assignment framework for heterogeneous underwater robot systems based on task value is proposed to address the task assignment problem with multiple types of tasks in emergency and non-emergency task scenarios.AUV can achieve fast access to tasks but in a high cost,while UG can achieve low-cost access to tasks with poor timeliness.Therefore,the adopted heterogeneous underwater robot system provides a highly matched collection method for different characteristic tasks,fully leveraging the advantages of heterogeneous underwater vehicles.A task value function is proposed to measure task value based on its importance,data volume,and timeliness.The calculated task value determines different access methods: Low value tasks are accessed through UG.An adaptive K-means clustering algorithm is proposed by adaptively determining the initial k-value based on the value of the task region.The motion regions are selected for UG according to the proposed cost-utility constraint principle to achieve low-cost access to tasks;High value tasks are accessed through the AUV system.A three-layer ESOM(Expanded SOM)network is proposed to achieve task allocation and path replanning.An expanded competitive learning mechanism is proposed in the expanded network to address the shortcomings of path intersection and self intersection,and thus achieve high cost efficiency access to tasks.Simulations were conducted in different scale task scenarios to verify the superior performance of the proposed task assignment framework for heterogeneous underwater robot systems.