Grounding Computer Use Agents on Human Demonstrations
This addresses the problem of building reliable computer-use agents for desktop environments, offering a significant improvement in data efficiency and performance, though it is incremental in advancing existing methods.
The authors tackled the lack of high-quality desktop grounding data for computer-use agents by introducing GroundCUA, a dataset with 56K screenshots and 3.56M annotations, and used it to train GroundNext models that achieve state-of-the-art results on benchmarks with less than one-tenth the training data of prior work.
Building reliable computer-use agents requires grounding: accurately connecting natural language instructions to the correct on-screen elements. While large datasets exist for web and mobile interactions, high-quality resources for desktop environments are limited. To address this gap, we introduce GroundCUA, a large-scale desktop grounding dataset built from expert human demonstrations. It covers 87 applications across 12 categories and includes 56K screenshots, with every on-screen element carefully annotated for a total of over 3.56M human-verified annotations. From these demonstrations, we generate diverse instructions that capture a wide range of real-world tasks, providing high-quality data for model training. Using GroundCUA, we develop the GroundNext family of models that map instructions to their target UI elements. At both 3B and 7B scales, GroundNext achieves state-of-the-art results across five benchmarks using supervised fine-tuning, while requiring less than one-tenth the training data of prior work. Reinforcement learning post-training further improves performance, and when evaluated in an agentic setting on the OSWorld benchmark using o3 as planner, GroundNext attains comparable or superior results to models trained with substantially more data,. These results demonstrate the critical role of high-quality, expert-driven datasets in advancing general-purpose computer-use agents.