LaLa-World: Scaling Latent World Models with Natural Language Actions

One-line takeaway

NAVER AI Lab's LaLa-World is a JEPA-style latent world model with a language-conditioned interface: describing both numerical actions and visual changes in language lets one future predictor train jointly on simulation, robot trajectories and broader image/video data, for planning in imagination and policy fine-tuning.

NAVER AI Lab 的 LaLa-World 是用自然语言描述动作的 JEPA 式潜在世界模型：把数值动作和画面变化都写成语言，从而用一个未来预测器联合学习仿真、机器人轨迹和更广泛的图像视频数据，用于“想象中规划”和下游策略微调。