汽车 None 对话罗剑岚：把机器人“部署”本身变成训练的一部分 作者 | 华卫 具身智能过去两年始终无法回避的核心矛盾：模型越来越强，但机器人仍然很“笨拙”。真实部署不是一个固定测试集，机器人进入商店、家庭和工作空间后，会不断遇到预训练数据没有覆盖的新物体、新摆放、新指令偏好和长尾失败。原因并不复杂，长 LWD reward 问题 2026-05-29 华卫
开源推荐 None GitHub 热门项目：RLHF 奖励建模 GitHub项目：RLHF-Reward-Modeling 仓库地址：https://github。com/RLHFlow/RLHF-Reward-Modeling Stars：1531 | 作者：RLHFlow 项目描述：Recipes the reward model 2026-05-24 GitHub Trending
智能AI None HealthCraft：急诊医学的强化学习安全环境 21496v1 Announce Type: new Abstract: Frontier language models are being deployed into clinical workflows faster than the and the safety 2026-05-23 Brandon Dent
智能AI None 超越模式搜索强化学习：扩散语言模型的轨迹平衡后训练 13935v1 Announce Type: new Abstract: Diffusion language models are a promising alternative to autoregressive models, yet the models reward 2026-05-15 Saba Ahmadi, Prasanna Parthasarathi, Yufei Cui
智能AI None $\xi$-DPO：通过比率奖励保证金直接偏好优化 arXiv:2605.10981v1 Announce Type: new Abstract: Reference-free preference optimization has emerged as an efficient alter the and reward 2026-05-13 Zhengyuan Fan, Zhonghua Wu, Yuxuan Du, Qun Chen
智能AI None 自动评分作为奖励：从隐式偏好到显式多模式生成标准 08354v1 Announce Type: new Abstract: Aligning multimodal generative models with human preferences demands reward signals that into reward 2026-05-12 Juanxi Tian, Fengyuan Liu, Jiaming Han, Yilei Jiang, Yongliang Wu, Yesheng Liu, Haodong Li, Furong Xu, Wanhua Li