Same Tasks, Different Apps: Why Mobile GUI Agents Fail to Generalize?
AnyAppBench is introduced, a category-controlled live Android benchmark that evaluates cross-application generalization while keeping the user goal fixed, and it finds that, across 13 agents, success on the original application does not transfer reliably to new applications with the same goal.
T. Trần, N. Koh, Daiki E. Matsunaga et al.
· 0 citations