Interactive Visual Gym: Closed-Loop Benchmark for Visual Interaction Models
arXiv — cs.CV metadata · Chenhao Zheng, Zixian Ma, Kai Li, Kanchana Ranasinghe, Jieyu Zhang, Artemis Panagopoulou, Weikai Huang, Jianxu Shangguan, Yuanyuan Tan, Silvio Savarese, Juan Carlos Niebles, Ranjay Krishna, Ran Xu, Honglu Zhou · Published on Readzer
Summary
Researchers introduced Interactive Visual Gym, a framework for closed-loop evaluation and training of visual interaction models using simulated task environments and human users. The benchmark IVG-Bench includes 315 scenarios across cooking, household manipulation, and mobile-interface domains. State-of-the-art models showed limitations, with the best (Gemini 4 Argon) completing only 53.4% of tasks within time limits, and a human study confirmed 78.5% agreement with the proposed interaction-quality metric.
This is a reviewed summary. Read the complete article at the original source.
Why it matters by industry
- Artificial intelligence
This work addresses a critical gap in AI evaluation by introducing closed-loop benchmarks for visual interaction models, which is directly relevant to advancing embodied intelligence and agentic systems in real-world tasks.