Research on AI Agent-Based Method for Automated Terminal Testing

With the complexity of terminal device functions, traditional testing methods face significant challenges, including limited operational flexibility and a high dependence on manual intervention. Therefore, the intelligent and flexible automated terminal inspection methods is becoming more necessary. In this paper, an automated testing method based on artificial intelligence (AI) agent with multimodal large language model (MLLM), leveraging the open-source Mobile-Agent-v2 framework, for mobile terminals is proposed. To enhance the generalizability and correctness of the proposed method in terminal testing scenarios, three key improvements are introduced: a) designing specialized operations for handling complex tasks, b) introducing a voting mechanism to improve decision stability, and c) optimizing prompts and external knowledge to enhance task comprehension. The proposed method is applied to terminal devices from multiple brands and validated using 32 real test cases provided by the telecom operator. Experimental results demonstrate that our proposed method achieves 87.5% task execution accuracy without manual intervention, while also exhibiting strong cross-device compatibility.