A Multi-Agent Model for Automatic Test Scheme Generation via Experience Interaction and 2D-Simulation Evaluation

With the rapid development of maritime intelligent systems and equipment, it has become increasingly urgent to effectively test the intelligence level and collaborative capabilities of these systems and devices. Currently, maritime intelligent systems and equipment testing is primarily conducted manually, involving analyzing the requirements for testing, generating test plans, and evaluating performance item by item. However, this manual approach faces challenges such as time-consuming and labor-intensive scheme planning, and overly simplistic test scenarios. Therefore, we propose a multi-agent model to automatically generate test schemes via Experience Interaction and 2D-simulation evaluation (MAEI-2D). MAEI-2D is designed to enable the automatic generation and optimization of test schemes for maritime systems and equipment by integrating large-scale task understanding, multi-agent collaboration, and two-dimensional simulation-based evaluation. It includes three agents, which perform generation, simulation, and evaluation, respectively. To improve the effectiveness of derivation from test description, an LLM-driven reasoning mechanism is introduced through natural language prompts. Experimental results on test scheme generation for maritime intelligent equipment demonstrate the performance of MAEI-2D.