StarCraft II AI combines fast actions with language model planning
JEV-Star: Fast, Low-Cost StarCraft II Control with Language-Model Planning
Computer Science and Game Theory
Summary
Playing StarCraft II well requires both quick decisions and longer plans. The authors created a system called JEV-Star that mixes fast automatic moves with slower, big-picture planning using a language AI called GPT-6. This combined system wins against strong in-game opponents while keeping decisions fast and costing little money. Their tests show it does better than just the fast automatic moves alone.
What this means in practice
- •For game developers: Integrate language-model planning with fast action selection to improve real-time strategy game AI opponents.
- •For robotics control teams: Use a combination of fast reactive controls and slower contextual planning to improve complex multi-agent coordination.
Authors
Weiyu Ma, Liangbing Zhao, Yongcheng Zeng, Jian Zhao
Abstract
We present JEV-Star, a StarCraft II controller that defeats the strongest non-cheating built-in AI, Lv7, by combining fast JEV action selection with persistent GPT-6 planning. The combined system wins four full games at Lv5--Lv7, including two Lv7 victories with different seeds, while retaining a median JEV response time of 0.422 seconds. Mean estimated model cost across these games is USD~3.71 per game: USD~0.15 for JEV and USD~3.56 for GPT-6. We compare this system with an initial JEV-only controller in full-game macro control and multi-unit micromanagement. The standalone controller reaches a 20-minute limit against Lv2 without expanding. Across 35 battle maps with three episodes per map and controller, the combined system raises mean enemy elimination from 16.50\% to 37.69\% and wins from 3 to 7 out of 105. Replay frames and decision logs document resource reservation, persistent economic goals, and stable army objectives in successful full games. The results demonstrate a practical division between inexpensive, subsecond decisions and longer-horizon planning. The comparison evaluates complete systems; concurrent interface improvements mean that planning's contribution is not isolated by a controlled ablation. Code is available at https://github.com/sc2musa/Jev_Star