Unsupervised learning based multi-AUV task allocation in 3-D underwater environments with ocean currents and obstacles

A current and obstacle adaptive self-organizing map (COA-SOM) algorithm is proposed for solving the challenges of uneven task distribution, high energy consumption, and slow convergence in multi-task assignment algorithms for Autonomous Underwater Vehicles (AUVs) in complex marine environments. Firstly, the competitive rule of SOM algorithm has been improved by incorporating the obstacle avoidance distance and ocean current influences factors. This improvement allows COA-SOM to select the optimal winner neuron in complex marine environments. Secondly, during the neuron weight update phase, an environmental influence function is introduced to enhance SOM algorithm’s adaptability to its environment. This environmental influence function employs the artificial potential field method to assess the impact of obstacle repulsion, current influence, and attraction of the target point on each weight update. Additionally, to expedite the algorithm’s convergence, an S-shaped dynamic acceleration factor is incorporated. This factor could expedite the update speed of neurons when obstacles are absent. Finally, the effectiveness of the COA-SOM algorithm is evaluated under different task sizes through comparisons with the SOM, PSO, DLBSOM, and GA algorithms in simulated and real ocean environments.