用 8 个 token 做规划：面向潜空间世界模型的紧凑离散分词器
Planning in 8 Tokens: A Compact Discrete Tokenizer for Latent World Model
把观测压缩到 8 个离散 token 的潜空间表示，可直接替换视觉世界模型中的重型分词器，使基于模型的规划在实时控制下变得可行；做视觉世界模型或 VLA 的读者可以借鉴其压缩比与规划保真度之间的权衡设计。
原文依据 compresses each observation into as few as 8 tokens, drastically reducing computational cost while preserving essential information for planning
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论世界模型提供了一个强大的框架，能够以动作或指令为条件模拟环境动态，从而支撑动作规划或策略学习等下游任务。近期方法将世界模型用作学习得到的模拟器，但将其应用于决策时规划时，计算开销仍然过高，难以满足实时控制的需求。关键瓶颈在于潜变量表示：传统分词器将每个观测编码为数百个 token，使规划既缓慢又耗费资源。为解决这一问题，我们提出 CompACT，一种离散分词器，可将每个观测压缩到少至 8 个 token，在保留规划所需关键信息的同时大幅降低计算成本。搭载 CompACT 分词器的动作条件世界模型在取得有竞争力的规划性能的同时，规划速度提升数个数量级，为世界模型的实际部署迈出了切实的一步。
英文原文摘要 954 字符
World models provide a powerful framework for simulating environment dynamics conditioned on actions or instructions, enabling downstream tasks such as action planning or policy learning.Recent approaches leverage world models as learned simulators, but its application to decision-time planning remains computationally prohibitive for real-time control.A key bottleneck lies in latent representations: conventional tokenizers encode each observation into hundreds of tokens, making planning both slow and resource-intensive.To address this, we propose CompACT, a discrete tokenizer that compresses each observation into as few as 8 tokens, drastically reducing computational cost while preserving essential information for planning.An action-conditioned world model that occupies CompACT tokenizer achieves competitive planning performance with orders-of-magnitude faster planning, offering a practical step toward real-world deployment of world models.