Researchers describe an online dense reinforcement learning approach that trains a planning-grounding agent for UI navigation on live iOS apps, targeting the high inference costs that make frontier vision-language models impractical for everyday use. The work argues that smaller models trained via supervised fine-tuning or offline reinforcement learning tend to memorize static trajectories rather than generalize.
While frontier vision-language models demonstrate strong capabilities in user interface (UI) navigation, their high inference costs prohibit everyday use. Conversely, smaller models trained via supervised fine-tuning (SFT) or offline reinforcement learning often memorize static trajectories, leading
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.