腾讯 的 类 mercor 的平台 experthub.tencent.com/?refer…
注册即送 100元 + 20元邀请费用,今日截止。
通过如上链接注册的朋友请私我并提供你的 uid , 回头我给你转20 。
1
894
grok回答里提到的这种迭代改进的方式,一方面,用户隐私风险(用户问的问题,都无条件成为它产品迭代的一部分,偷 prompt);
另一方面,从每日 pb 级别日志中洗出高价值任务和长程任务,本身就是 ROI 极低的事情。国内 中转站黑灰产才这么干。
我认为最合理的解法还是 类似 字节 xpert ,腾讯 experthub 这样的 共赢方式,尊重知识产权 ...
在构建专家网络这一个方面,字节 和 腾讯 又走在了前面
Replying to @seclink
还在运营。官网mulerun.com仍可访问,虽团队已并入钉钉悟空并整合进千问办公,但产品持续服务。
它不消费传统公开大数据集,而是通过三层自进化引擎持续改进:任务层记忆个人工作流偏好,领域层主动获取专业技能,社区层共享匿名化成功模式与集体智慧。每次任务执行都反馈优化Agent,私人数据严格隔离,只共享结构知识。
1
836
国外都在抢占高地用 AI 制作 PCB板 ,
国内? 还在清洗 中转站的地沟油 ....
洗 10pb 数据都洗不出高价值的数据来 ...
而且,不重视客户知识产权和客户的隐私 ...
Grok and Opus given the same task: design a USB temperature sensor PCB in KiCad.
Left: Grok 4.7 | Right: Claude Opus 5.5
Grok 4.7 went with a STM32 MCU + MCP9808, 52×24 mm PCB
Claude Opus 5.5 did a MCP2221A USB-I²C bridge + MCP9808, 30×26 mm
Which would you send to fab?
1
3
830
帖子视频标注主要用Meta Sapiens2(外视角全身308关键点姿态估计)和NVIDIA EgoHand(第一人称手部3D重建,基于SOMA-Hand参数模型),检测用RT-DETR。均有开源:Sapiens2在github.com/facebookresearch/…;EgoHand推理在github.com/NVIDIA/IsaacCaptu…(isaaccapture_egopost),模型在HF。
Y11 retweeted
NVIDIA just released a PCB assembly OpenUSD dataset on Hugging Face
Two board designs with 702 placed components, ready for synthetic AOI data generation and 3D inspection research.
huggingface.co/datasets/nvid…
Y11 retweeted
You can buy 100K hours of first-person video and barely improve a Physical AI model's performance on the robot.
Human egocentric data collection has grown quickly in the last few years. From where I stand, the excitement is grounded in its potential for quick task adaptation, generalization to novel objects, and robotics tasks that only unlock with hand manipulation.
I'm still putting the full picture together, but a few patterns are becoming clear.
Promising results show that downstream robot performance improves with more human data, especially in novel environments and with objects unseen during training. Both are exactly the conditions robot deployment companies care about.
There are two types of egocentric data collected today: passive video and deliberate demonstration.
Passive video is simpler to collect, but rarely action-labelled. It is captured with less intent, which typically results in unplanned head motion and long stretches of idle, irrelevant activity.
Deliberate egocentric demonstrations are collected with specialized hardware that captures video alongside hand positions, which can be mapped onto the robot's actual hand. These datasets are closer to what a buyer would pay for if the target use case is low-shot manipulation adaptation.
Both types of egocentric data have a role, because each teaches a model something different.
Ultimately, large volumes of POV footage are available, but what matters is how much they improve Physical AI model performance on robot tasks.
Running multimodal analysis on petabyte-scale egocentric data to determine its quality means ingesting and processing it frame by frame.
We heard others were running into challenges doing so, which motivated us to build Pareto (lnkd.in/ggBJhyqU) as open source.
If you're an egocentric data company, send us your gnarliest dataset and we'll show you which parts of it are robot-relevant.
Y11 retweeted
We’re excited to announce RoboMoCap.
The largest egocentric + truly ground-truth whole-body dataset for humanoids.
We think this is an important first step towards bootstrapping whole-body generalization from scalable first-person POV data.
With @BitRobotNetwork @NokovMocap
🚀 Introducing RoboMoCap!
We are releasing the largest locomotion-centric egocentric vision × optical motion capture dataset for humanoid robot learning, joint effort with @GroundedSI @BitRobotNetwork @NokovMocap
📊200+ hours of recordings
🏃 290+ diverse actions
👓 Egocentric Vision
🎥 120 Hz optical motion capture with ~10 mm tracking accuracy
🌐 bitrobot-foundation.github.i…
🧵👇
Y11 retweeted
Can a human motion prior help egocentric SLAM, and can SLAM in turn help egocentric motion reconstruction?
Presenting EgoMo3R, our #NeurIPS2026 paper on joint 3D motion and scene reconstruction from egocentric video.
chaitanya100100.github.io/Eg…
Y11 retweeted
1/ Robots learn by watching people work. Here are 100 hrs of it, in 3D.
RoboCap LiDAR-100 is the largest egocentric LiDAR (QVGA) dataset to date.
Real kitchen / workshop shifts were captured with @RoboSenseLiDAR, with RGB, QVGA depth and IMU in sync. Explore the data↓
Y11 retweeted
9TB of egocentric video open sourced on @huggingface
@eidon_ai is winding down and decided to open-sourced their data. 1,274 hours of first-person housework video, paired with 7-point IMU arm tracking. 13,451 recordings from 27 homes. CC-BY-4.0.
huggingface.co/datasets/eido…
数据集,可以下载看看
Today, we're releasing EgoCrowd-500H, a new dataset accompanying our @corl_conf 2026 research on quality failures in crowdsourced egocentric video.
• 300 stereo clips
• 2.5 hours of footage
• 50 human-reviewed clips
Out now on @huggingface.
413
这篇文章提供的是数据集还是提供什么方法?还是提供了服务
Today we’re releasing Perceptron Egocentric 1.0: The frontier for egocentric annotations, now through a self-service API.
• Fine-grained subtasks matching Astra quality at ~1/8 the cost
• State-of-the-Art 2D and 3D hand tracking
• Individual timelines for every action on each hand and robot gripper
544
















