Research map / Multi-agent systems and games / LLM agents
UINavBench: A Framework for Comprehensive Evaluation of Interactive Digital Agents
Harsh Agrawal, Eldon Schoop, Xinlei Pan, Anuj Mahajan, Ari Seff, Di Feng, Ruijia Cheng, Andres Romero Mier Y Teran, Esteban Gomez, Abhishek Sundararajan, Forrest Huang, Amanda Swearngin and 3 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add simplified text representation of detected UI elements to set-of-marks prompting (ablation) | ReACT agent (GPT-4o) · UINavBench | Episode Success: 25.8% -> 31.8% | Improved |
| add Reflexion self-critique with a VLM critic (ablation) | ReACT + Text agent (GPT-4o) · UINavBench | Episode Success: 31.8% -> 40.5% | Improved |
| replace heuristic SoM grounding with the UGround grounding model (ablation) | Reflexion + Text agent · UINavBench | Episode Success: 40.5% -> 33.5% | Hurt |
| use end-to-end UI-TARS-72B instead of modular GPT-4o planner | UINavBench agents · UINavBench | Episode Success: 40.5% -> 11% | Hurt |
| increase task complexity from simple to complex with deep navigation | Reflexion + Text · UINavBench | success: 11/14 (79%) -> 4/36 (11%) | Hurt |
| require visual understanding instead of text-only perception | Reflexion + Text · UINavBench | Episode Success: 80% -> 7.7% | Hurt |
| require world knowledge outside the environment | Reflexion + Text · UINavBench | Episode Success: 47.7% -> 13.3% | Hurt |
| Reflexion on data entry tasks (ablation) | Reflexion + Text · UINavBench data entry tasks | Episode Success: 10.6% -> 8.5% | No clear effect |
Closest papers by meaning
- SPA-BENCH: A COMPREHENSIVE BENCHMARK FOR SMARTPHONE AGENT EVALUATION (ICLR 2025 · similarity 0.92)
- VenusBench-Mobile: A Challenging and User-Centric Benchmark for Mobile GUI Agents with Capability Diagnostics (ICML 2026 · similarity 0.91)
- AndroidWorld: A Dynamic Benchmarking Environment for Autonomous Agents (ICLR 2025 · similarity 0.90)
- REAL: Benchmarking Autonomous Agents on Deterministic Simulations of Real Websites (NeurIPS 2025 · similarity 0.90)
- MMBench-GUI: A Unified Hierarchical Evaluation Framework for Multi-Platform GUI Agents (CVPR 2026 · similarity 0.90)
- ProactiveMobile: A Comprehensive Benchmark for Boosting Proactive Intelligence On Mobile Devices (CVPR 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
