From Tapping to Hopping: Augmenting Mobile GUI Agents with App-Native Deeplinks
Yuchen Sun, Yue Wang and colleagues at Shanghai Jiao Tong University and Tongyi Lab, Alibaba Group train mobile GUI agents to call verified app deeplinks for navigation and fall back to taps and swipes for everything else.
Ask this paper
Verified deeplink catalog. Candidate deeplinks are found by static analysis of apps, validated on real devices and paired with a description of the screen each one opens.
GUI-Hopper. A Qwen3.5-based agent trained on hybrid trajectories that mix deeplink jumps with GUI actions.
MobileWorld results. At 9B, success rises from 54.4% for the GUI-only trained agent to 63.5% with deeplinks; at 4B, from 52.7% to 60.1%. Average steps drop 15.3-21.3%.
Gains without deeplinks. Evaluated with GUI actions only, the hybrid-trained model still beats the GUI-only agent at every scale (56.7% vs 54.4% at 9B), which the authors attribute to better intermediate-goal planning.
Real devices. The gains hold on commercial apps running on physical phones.
Abstract
Mobile GUI agents complete tasks using GUI actions like taps and swipes. These actions are broadly applicable across applications, but reaching a navigation interface. A single deeplink call can replace a sequence of screen-by-screen GUI actions. We therefore introduce hybrid interaction, using deeplinks for direct navigation and GUI actions for other on-screen operations and fallback. To enable this, we discover candidate deeplinks through static analysis, validate them on real devices, and describe their observed landing screens. This process creates a verified and grounded deeplink catalog that pairs each working deeplink with a description of its landing screen. Using this catalog, we introduce GUI-Hopper, a improves task success in commercial applications on real devices, further demonstrating the benefits of hybrid interaction.