[论文] AgentGarten: Code Worlds for Evolving Agents
研究领域: CV 作者: Jiawei Chi, Shangchen Miao, Zhiyuan Shi, Kailu Wu, Hanyang Wang, Weiliang Chen, Qiyu Dai, Jinshan Ren, Jun Gao, Mingsheng Long, Yueqi Duan, Jiangr…
论文概要
研究领域: CV 作者: Jiawei Chi, Shangchen Miao, Zhiyuan Shi, Kailu Wu, Hanyang Wang, Weiliang Chen, Qiyu Dai, Jinshan Ren, Jun Gao, Mingsheng Long, Yueqi Duan, Jiangran Lyu, Jialong Wu, Fangfu Liu 发布时间: 2026-10-08 arXiv: 2610.12374
中文摘要
交互虚拟世界让 agent 通过探索与交互学习,但 agent 能学到什么受限于练习环境的保真度(一致的状态、规则、动态)和真实感(符合真实世界视觉分布的观测),二者兼得且覆盖多样世界仍是瓶颈。我们提出 AgentGarten:将模拟器/游戏引擎与共享神经渲染器耦合来构建实时交互环境。模拟后端维护持久世界状态并执行程序定义的交互规则,渲染器通过通用接口从结构化条件生成视觉观测。训练神经渲染器时,我们将预训练视频模型适配到几何条件,用提出的 Adversarial Forcing 进行蒸馏并优化推理实现实时交互:Adversarial Forcing 通过精确回放使历史预填充可微,后续预测的损失会反向更新渲染器对先前观测的编码方式,并加入真实数据对抗监督提升视觉质量。在 AgentGarten 中,agent 通过视觉观测感知世界、实时交互,并将每轮经验蒸馏成 playbook 供后续 agent 继承和精炼。实证研究显示学习效率大幅提升:agent 仅用 4 轮学习即超过传统强化学习需要数百万轮的水平。由于新世界可以写成代码并通过同一接口渲染,环境可以随 agent 一起在数量和难度上扩展——这是迈向通过交互经验持续进化的 agent 的一步。
原文摘要
Interactive virtual worlds allow agents to learn through exploration and interaction. What agents can learn is bounded by the environments they practice in, which must be faithful, with consistent state, rules, and dynamics, and realistic, with observations that follow the real-world visual distributions. Achieving both across diverse worlds remains a bottleneck. We introduce AgentGarten, a framework that couples simulators and game engines with a shared neural renderer to build real-time interactive environments. Its simulation backends maintain persistent world state and execute program-defined interaction rules, while the renderer generates visual observations from structured conditions exported through a common interface. To build the neural renderer, we adapt a pretrained video model ...
*自动采集于 2026-10-12*
#论文 #arXiv #CV #小凯