MaP Framework Unifies GUI Navigation via Masked Trajectory Prediction
September 23, 2026
The MaP framework addresses data heterogeneity in GUI agents by modeling multi-turn interactions as trajectories. It uses component masking and prediction to harmonize divergent navigation tasks into a single optimization objective for better long-horizon planning.
HOW THIS AFFECTS YOU
●
builderThis method could improve the reliability of agents performing complex, multi-step software workflows.
●
researcherYou can use this unified objective to stabilize training across heterogeneous GUI datasets.