New benchmark PokeGym tests AI agents in Pokémon Legends: Z-A

Researchers have introduced PokeGym, a new benchmark environment built on the game Pokémon Legends: Z-A, designed to test vision-language agents' ability to adapt and learn across multiple game episodes without direct access to game states. To address the challenges of this environment, they also developed G-EvoMAC, a graph-guided framework that synergistically optimizes visual perception, strategy, and action selection. Experiments demonstrated that G-EvoMAC achieved a 60.18% success rate on PokeGym, significantly outperforming existing methods. AI

IMPACT Introduces a new benchmark for evaluating multimodal AI agents in complex, visually-driven environments.

RANK_REASON Academic paper introducing a new benchmark and framework. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• G-EvoMAC
• Gotit.pub
• Hugging Face
• PokeGym
• Pokémon Legends: Z-A
• ScienceCast
• Ye Huang

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →