MAC-v2：从像素出发的长时程基于模型价值扩展方法
MAC-v2: Long-horizon model-based value expansion from pixels
一篇离线强化学习新论文，MAC-v2 用潜在世界模型加动作分块，在 LIBERO-10 这些视觉基准上跑赢了 model-free 和 model-based 基线，做 RL 的可以看看。
论文提出 MAC-v2，用于像素输入的离线强化学习。方法核心是用不依赖重建的潜在世界模型配合动作分块（action chunking），减少长时程想象轨迹中的递归预测误差累积。在视觉 OGBench、MMBench2、Robomimic 和 LIBERO-10 等基准上，MAC-v2 超越了 model-free 和 model-based 离线 RL 基线。消融分析显示长时程价值扩展、动作分块和潜在学习目标三者缺一不可。
MAC-v2: Long-horizon model-based value expansion from pixels
We study whether \emph{long-horizon model-based value expansion} can serve as an effective framework in pixel-based offline reinforcement learning (RL). Long-horizon model-based value expansion from pixels is substantially harder than state-based environments, as the agent must learn both the representation and its dynamics, and errors from either can accumulate recursively over imagined trajectories. We introduce MAC-v2, which enables long-horizon model-based expansion from pixels by training a reconstruction-free latent world model with action chunking. By learning representations from prediction and control objectives, MAC-v2 does not need to encode control-irrelevant visual details that can hinder the dynamics prediction. Action chunking further mitigates the prediction error by reducing the number of recursive model predictions required for value expansion. Across diverse visual benchmarks including visual OGBench, MMBench2, Robomimic, and LIBERO-10, MAC-v2 achieves strong performance against model-free and model-based offline RL baselines. Our analyses show that long-horizon model-based value expansion is critical for the performance of MAC-v2, and that both action chunking and latent learning objectives are essential for reliable long-horizon latent imagination.