TL;DR
Existing benchmarks for vision-language model (VLM) agents in games are limited, focusing on single attempts and lacking unified evaluation methods. OmniGameArena was developed as a real-time benchmark featuring twelve new Unreal Engine 5 games to address these issues.
✦ Why It Matters
Engineers can leverage OmniGameArena to evaluate and improve VLM agents in diverse gaming scenarios effectively.
Key Takeaways
How It Works
OmniGameArena standardizes the evaluation of VLM agents by providing a unified action interface across various game types. The Improvement Dynamics Curve (IDC) allows agents to iteratively refine their skills through a language model that reflects on their performance, enabling a deeper analysis of how agents adapt and improve over time.
Related