返回 2026-08-18 编辑版
发现与解读中文优先 · EN 原文机器人

Moving Atoms (YC S26) 发布:Atom 1 用视频世界模型训练机器人

我们正在利用在互联网规模视频上训练的世界模型来解决机器人问题:这些模型能够准确预测物理交互并以机器人动作为条件。我们的第一个模型 Atom 1 已经名列前茅DeepMind的物理IQ超越标杆NVIDIA的宇宙3.

查看英文原文

We're solving robotics with world models trained on internet-scale video: models that accurately predict physical interactions and are conditioned on robot actions. Our first model, Atom 1, already tops DeepMind’s Physics IQ benchmark - beating NVIDIA's Cosmos 3.

We're solving robotics with world models trained on internet-scale video: models that accurately predict physical interactions and are conditioned on robot actions. Our first model, Atom 1, already tops DeepMind’s Physics IQ benchmark - beating NVIDIA's Cosmos 3.

Agent 正从屏幕内推理走向持续感知与设备操作

为什么值得看

持续感知、动作条件和真实设备接口会把 Agent 的价值与风险同时放大。产品判断需要新增延迟、动作可逆性、权限边界和现实环境失败恢复,而不只是语言回答质量。

最强反方

Atom 1 超过既有物理 benchmark 与 MOSS-VL 的能力来自项目方或论文传播,两个手机项目来自作者提交;这些材料没有提供统一硬件条件、误操作率或安全评测。

原始出现与传播对照

1 条出现 · 本次 AIHOT 未精确命中

发现与解读

我们正在利用在互联网规模视频上训练的世界模型来解决机器人问题:这些模型能够准确预测物理交互并以机器人动作为条件。我们的第一个模型 Atom 1 已经名列前茅DeepMind的物理IQ超越标杆NVIDIA的宇宙3.

我们正在利用在互联网规模视频上训练的世界模型来解决机器人问题:这些模型能够准确预测物理交互并以机器人动作为条件。我们的第一个模型 Atom 1 已经名列前茅DeepMind的物理IQ超越标杆NVIDIA的宇宙3.

查看英文原文

We're solving robotics with world models trained on internet-scale video: models that accurately predict physical interactions and are conditioned on robot actions. Our first model, Atom 1, already tops DeepMind’s Physics IQ benchmark - beating NVIDIA's Cosmos 3.

We're solving robotics with world models trained on internet-scale video: models that accurately predict physical interactions and are conditioned on robot actions. Our first model, Atom 1, already tops DeepMind’s Physics IQ benchmark - beating NVIDIA's Cosmos 3.

直接发布标题Moving Atoms (YC S26)@MovingAtomsLab打开该层出处 ↗

We're solving robotics with world models trained on internet-scale video: models that accurately predict physical interactions and are conditioned on robot actions. Our first model, Atom 1, already tops DeepMind’s Physics IQ benchmark - beating NVIDIA's Cosmos 3.

来源
X @ycombinator · x-ycombinator
证据角色
x_project_network
时间状态
known
用户判决
未判
查看该来源原文raw/x-ycombinator.json456f7ad15e343498ae9ba68155f36302654d2db5ffc7c9ab0dc22168995f2ead
Cluster IDevt_1932bfa34fc2d965e769b75d精确去重依据canonical_url事件状态未判