🤖
RL Token：用视觉-语言-动作模型启动在线强化学习 深度解读
Physical Intelligence 最新研究：通过从预训练 VLA 模型中提取紧凑的 RL Token 表示，RLT 方法能够在仅几小时机器人实践中高效在线微调大型 VLA，实现成功率提升高达 3 倍。
Read More →Technical deep dives and philosophical reflections from the XRollout community. Exploring the future of open-source robot learning.
Physical Intelligence 最新研究：通过从预训练 VLA 模型中提取紧凑的 RL Token 表示，RLT 方法能够在仅几小时机器人实践中高效在线微调大型 VLA，实现成功率提升高达 3 倍。
Read More →Our mission - why we started XRollout and what we believe. Robotics should be open, accessible, and community-driven.
Read More →Our complete learning philosophy — four pillars of robot acquisition and the hierarchical data pyramid.
Read More →From prefrontal cortex vs basal ganglia to why both vision and language are indispensable in VLA.
Read More →How community feedback closes the data loop and makes robots smarter in real-world scenarios.
Read More →Incentivizing contribution with a credit system where you earn by contributing and redeem for platform resources.
Read More →How memory architectures inspired by human cognition enable robots to make better sequential decisions by learning from past experience.
Read More →Why the memory bottleneck isn't model architecture—it's data. How SLAM provides the spatio-temporal structure memory models need, and three business opportunities.
Read More →SLAM isn't just a robotics algorithm—it's a theory of memory. Three levels of structural isomorphism between SLAM components and human memory systems.
Read More →清华团队提出质疑：视频预测的主要价值来自训练阶段的表示学习，而非测试阶段的显式未来想象。速度提升 4 倍，性能仅下降 2-3%。
Read More →Physical Intelligence 最新研究：通过短期视频编码 + 长期语言记忆，让 VLA 完成长达 15 分钟的复杂任务，如整理厨房、制作三明治。
Read More →Moonshot AI 最新研究：揭示 Transformer Attention 的双重残差效应——低秩性与共线性，提出 Persistency Map 可视化工具与三种优化算法。
Read More →全面梳理 LIBERO、RLBench、CALVIN、MetaWorld、RoboSuite 等主流机器人学习 benchmark，详解各基准的特点、任务设计、评估指标及选择指南。
Read More →从原始演示数据到可执行技能配方：多模态分析 → Scene Understanding → Affordance Map → 层次化记忆 → SLAM 几何锚定 → Skill Textbook 完整架构。
Read More →