让AI在GTA里考驾照
游戏一直是AI的练车场，但没人能给「任务有没有完成」自动打分——大多得靠人盯着看。这篇把GTA V的主线任务改造成103个可自动判定的考题，外加47个基于游戏画面的射击狙击题，还搭了个服务器：AI只看截图、只发键鼠操作，引擎自己检查任务成没成。六个前沿大模型来考，最好的配置总准确率70.9%，但空中导航、甩追兵、拦截这几项依然吃力。更有意思的是后半段：他们做了个任务生成器，在固定模板里随机采样参数，批量造出带自动奖励和终止信号的训练环境，拿它做强化学习，一个27B模型在步行导航上练完，不光本类任务变强，还迁移到了没见过的任务类型，原生GTA V任务成功率从15%涨到23%。它不是你明天能用上的东西——但「用3A游戏当通用智能体的考场兼训练场」这条路，被这篇往前推了一格。
📄 原文摘要(英文)
AAA games offer a rich testbed for evaluating agent perception, reasoning, and control. However, evaluating whether agents can complete tasks in AAA games is challenging and often human-annotated because task outcomes are often difficult to verify automatically, let alone applying it for Reinforcement Learning (RL). We introduce GTAGym, an environment and framework for evaluation and RL on GTA V tasks. Its evaluation benchmark, GTAVBench, derived from GTA V main-story missions, contains 103 online tasks reconstructed across nine task families, alongside 47 image-based shooting and sniping tasks using GTA V imagery. A web server provides access to the online environment and evaluation interface, supporting parallel testing through screenshots and keyboard-mouse actions with automatic engine-state outcome checks. We evaluate six frontier LLMs on GTAVBench with textual task instructions and on its GTAVBench-AGI branch, where instructions appear only in the first screenshot and a harness automatically compresses interaction history. The best configuration achieves 70.9% overall accuracy in GTAVBench-AGI, yet aerial navigation, evasion and interception remain difficult. GTAGym also provides a scalable generator that samples task parameters within user-specified ranges and option sets under fixed templates, creating batches of 3D and image-based shooting environments for RL with automatic reward and termination signals. LoRA RL of Qwen3.8-27B on generated on-foot navigation tasks improves the in-domain performance and transfers to out-of-domain task categories, and native GTA V tasks from 15% to 23%.