VeriCUA: Learning Proactive Verification for Reliable GUI Agent Rewards
Reinforcement learning for computer-use agents increasingly relies on proactive reward agents that interact with the post-execution GUI environment to verify what recorded screenshots cannot reveal.
ProofPaper ↗
Key points
- The reliability of such verification hinges on two coupled decisions: what to check and when to stop.
- We introduce VeriCUA, a proactive reward agent that learns its verification strategy from these single attempts.
- Online reinforcement learning optimizes the Judge for both verdict correctness and efficient Probe use, with returns normalized across terminal states.
- Trained only on OSWorld and evaluated without further training on MobileWorld and WebArena-Verified, VeriCUA achieves the highest reward agreement across all three benchmarks and lies on the agreement-cost Pareto frontier among proactive verifiers.
Sources (1)
- [1]VeriCUA: Learning Proactive Verification for Reliable GUI Agent RewardsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 02:17 PM
Reinforcement learning for computer-use agents increasingly relies on proactive reward agents that interact with the post-execution GUI environment to verify what recorded screenshots cannot reveal.
The reliability of such verification hinges on two coupled decisions: what to check and when to stop.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 19, 2026CUA-Lite: An Open Platform for Computer Use Agents
- Oct 19, 2026AI researchers stranded Claude on a virtual desert island. It started to “play.”
- Oct 19, 2026EmbodiedWorld-200K: Toward Embodied World Agents via Embodied-Planning Dataset and Interactive World Models
- Oct 19, 2026Agent World Model in the Wild: Computer-Use Agent Environment Synthesis
- Oct 18, 2026CoworkEnv: Synthesizing Training Environments for Long-Horizon Cowork Agents
- Oct 17, 2026HybridResearcher: Bridging the Structured-Unstructured Boundary for Deep Research Agents