D-GARA: A Dynamic Benchmarking Framework for GUI Agent Robustness in Real-World Anomalies

This article introduces D-GARA, a framework designed to evaluate the robustness of GUI agents in the face of real-world anomalies. The focus on dynamic benchmarking suggests an attempt to create a more realistic and challenging evaluation environment compared to static benchmarks. The use of 'real-world anomalies' implies the framework considers issues like unexpected UI changes, network latency, or other factors that can impact agent performance. The source being ArXiv indicates this is likely a research paper.