Search papers, labs, and topics across Lattice.
UI-Venus-2 is a versatile foundation GUI agent that enhances digital task automation across mobile, web, and desktop environments by implementing a closed-loop reasoning-action framework. The research addresses critical challenges in real-world deployment, including environment coverage, task construction, and reward verification, by expanding to over 170 multilingual apps and utilizing advanced evaluators for reliable training signals. Key findings indicate that the integration of safety-aware mechanisms significantly improves the agent's ability to execute complex tasks while maintaining control over consequential actions.
UI-Venus-2 achieves unprecedented environment coverage and task reliability, paving the way for dependable multimodal GUI agents in real-world applications.
Multimodal GUI agents have emerged as a promising paradigm for digital task automation, yet transitioning from benchmark-oriented models to dependable real-world applications remains challenging due to limited environment coverage, brittle task construction, and unreliable reward verification. In this work, we present UI-Venus-2, a general-purpose foundation GUI agent designed to operate across mobile, web, and desktop environments through a unified closed-loop reasoning-action framework. To bridge the gap toward practical deployment, we jointly scale three critical dimensions: (1) Environments, expanding coverage to more than 170 multilingual mobile apps and native desktop operating systems; (2) Tasks, employing a deep-research pipeline for function-grounded instruction generation; and (3) Verification, adopting trace-level and sample-level evaluators with visual keypoints and multi-model voting to ensure reliable RL signals for training. Furthermore, we integrate safety-aware mechanisms to ensure controlled execution of consequential actions. By offering a capable, efficient, and open-source foundation, UI-Venus-2 advances the field toward more generalizable, verifiable, and self-reflective agents for real-world applications.
