What's Missing in Screen-to-Action? Towards a UI-in-the-Loop Paradigm for Multimodal GUI Reasoning

arXiv:2604.06995 · cs.AI · Submitted 2026-04-08 · Read on arXiv

cs.AI

Submitted: 2026-04-08

Updated: 2026-10-02

Comments: Accepted by ACL 2026 Findings

Code: https://github.com/zjukg/UILoop

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Benchmark — 1× in this paper · explained in 2 episodes
• Existing Graphical User Interface — 1× in this paper
• Interpretability — 1× in this paper · explained in 15 episodes

Sources

• Qwen2.5-VL Technical Report
• The BrowserGym Ecosystem for Web Agent Research
• ARM-Thinker: Reinforcing Multimodal Generative Reward Models with Agentic Tool Use and Visual Reasoning
• NeuReasoner: Towards Explainable, Controllable, and Unified Reasoning via Mixture-of-Neurons
• Learning Human-Perceived Fakeness in AI-Generated Videos via Multimodal LLMs
• Temp-R1: A Unified Autonomous Agent for Complex Temporal KGQA via Reverse Curriculum Reinforcement Learning
• DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
• UniCorn: Towards Self-Improving Unified Multimodal Models through Self-Generated Supervision
• ReCreate: Reasoning and Creating Domain Agents Driven by Experience
• GUI Agents with Reinforcement Learning: Toward Digital Inhabitants
• OS Agents: A Survey on MLLM-based Agents for General Computing Devices Use
• GPT-4o System Card
• FoE: Forest of Errors Makes the First Solution the Best in Large Reasoning Models
• Video-MME-Logical: A Controlled Diagnostic Benchmark for Video Temporal-Logical Reasoning
• Last Layer Logits to Logic: Empowering LLMs with Logic-Consistent Structured Knowledge Reasoning
• Mapping Natural Language Instructions to Mobile UI Action Sequences
• Make LVLMs Focus: Context-Aware Attention Modulation for Better Multimodal In-Context Learning
• MMFineReason: Closing the Multimodal Reasoning Gap via Open Data-Centric Methods
• When Models Learn to Ask Why: Adaptive Causal Reasoning for Trustworthy Medical Vision-Language Models
• General365: Benchmarking General Reasoning in Large Language Models Across Diverse and Challenging Tasks

Related papers

• MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
• Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
• Agents' Last Exam
• The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
• MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
• Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems