跳到正文
arXiv cs.AI· Zipeng Qiu, Shusheng Xu, Jiale Li, Yukun Zhou, Chenyue Li, Wenshuang Li, Tongkai Yang, Binhang Yuan·· 10 小时前AI 评分48

离线思考、实时行动:通过代码实现自我改进的视觉游戏智能体

Think Offline, Act in Real Time: Self-Improving Visual Game Agents through Code

AI 导读

研究对比了两种实时视觉游戏中的自我改进策略:在 170 个 GameWorld 任务、每次最多 8 次一分钟执行下,"代码即策略"(模型在尝试间编写、测试并修订程序)达到 47.80% 平均进度,优于训练的 2B 控制器(30.90%),并在全部五个游戏类别中领先。在单个游戏上微调控制器权重收益有限。结果表明让模型离线思考、把所学存入实时运行的程序更有效。

正文 · 原文

View PDF HTML (experimental)

Abstract:Pre-trained models can read screens and reason about games, yet on an unfamiliar game, an agent still has to learn to play. We ask how an agent can self-improve on such a game from a few of its own attempts, in real-time visual games where the agent sees only the screen and the game keeps running while the agent thinks. We study two kinds of policy that can carry this improvement. In model as policy, a model selects actions during play; we adapt real-time chunking so that committed actions keep executing while the model reasons, and train a 2B controller for this interface. In code as policy, a model writes, tests, and revises a program between attempts, and the program acts during play. Across 170 GameWorld tasks, with at most eight one-minute executions per run, code as policy reaches 47.80% mean progress, against 30.90% for our trained controller, and leads in all five game categories. Improving the controller's weights on a single game, by contrast, yields only modest gains. These results favor letting a model think offline and storing what it learns in a program that acts in real time. Project repository: this https URL
Subjects: Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.24590 [cs.AI]
  (or arXiv:2611.24590v1 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2611.24590

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Zipeng Qiu [view email]
[v1] Wed, 18 Nov 2026 15:18:22 UTC (1,266 KB)

来源:arXiv cs.AI · arxiv.org