UMI-style manipulation data · SenseXperienceUMI 风格操作数据 · SenseXperience
Handheld gripper data for robot manipulation面向机器人操作的
手持夹爪数据
Collect manipulation demonstrations without a robot on site. The SenseXperience UMI Unit — a Universal Manipulation Interface–style handheld gripper — records camera, jaw angle, IMU and audio in sync, and EmbodiFlow turns every session into training-ready data.无需机器人在场，即可采集操作示范。SenseXperience UMI 单元是一款 UMI（Universal Manipulation Interface）风格的手持夹爪，同步记录相机、夹爪开合角度、IMU 与音频，再由 EmbodiFlow 把每一次采集转化为可训练数据。
- Weight重量single hand单手160 ± 5 g
- Gripper夹爪opening angle · max opening · max force开合角度 · 最大开口 · 最大夹持力60° · 155 mm · 5 kg
- RGB cameraRGB 相机170° FoV170° 视场角2048×1536 · 30 Hz
- Magnetic encoder磁编码器jaw angle, 1.0 ± 0.2°夹爪角度，精度 1.0 ± 0.2°240 Hz
- IMU±16 g · ±2000°/s±16 g · ±2000°/s240 Hz
- Microphone麦克风omnidirectional全向2 channels双声道
Why UMI-style data为什么是 UMI 风格数据
Manipulation demonstrations, no robot required无需机器人在场的操作示范
Hardware-agnostic capture such as UMI lets people demonstrate manipulation directly with a handheld gripper. For VLA models, Physical AI and world models, this embodiment-agnostic human data is becoming a critical training foundation.UMI 这类与硬件无关的采集方式，让人直接用手持夹爪完成操作示范。对 VLA 模型、物理 AI 与世界模型而言，这类与本体无关的人体数据正成为关键的训练基础。
Collect in the wild真实场景采集
Compact, easy-to-use hardware for scalable data collection in real homes and workplaces — no robot cell required.结构紧凑、上手简单，可在真实家庭与工作场所规模化采集，无需搭建机器人工位。
Gripper-native signals原生夹爪信号
The gripper’s jaw angle is measured by a magnetic encoder and recorded alongside the camera and IMU streams.夹爪开合角度由磁编码器测量，与相机、IMU 数据一同记录。
Diverse tasks, long tail多样任务，覆盖长尾
Human data covers diverse task types, complex manipulation strategies and unstructured environments — and scales for long-tail tasks.人体数据覆盖多样的任务类型、复杂的操作策略与非结构化环境，更易扩展到长尾任务。
What gets recorded采集内容
Every grasp, time-synchronized每一次抓取，时间同步记录
Gripper camera夹爪相机
Wide 170° RGB view of the interaction zone, 2048×1536 at 30 Hz.170° 广角 RGB 画面对准交互区域，2048×1536 @ 30 Hz。
Jaw angle夹爪开合
Magnetic encoder at 240 Hz with 1.0 ± 0.2° accuracy.磁编码器 240 Hz 采样，精度 1.0 ± 0.2°。
Motion运动
On-board IMU at 240 Hz: accelerometer ±16 g, gyroscope ±2000°/s.板载 IMU 240 Hz：加速度计 ±16 g，陀螺仪 ±2000°/s。
Audio音频
Two-channel omnidirectional microphone for contact and event cues.双声道全向麦克风，记录接触与事件声音线索。
Build the full rig组合完整采集系统
UMI, wrist and egocentric views — together or on their ownUMI、腕部与第一视角，可组合也可独立使用
SenseXperience’s baseline configuration combines four modules. Each can be deployed independently or configured together as one unified system.SenseXperience 基线配置由四个模块组成，每个模块都可以独立部署，也可组合成统一系统。
UMI Unit
Gripper acquisition夹爪数据采集
Handheld gripper with camera, encoder, IMU and microphone.集成相机、编码器、IMU 与麦克风的手持夹爪。
Wrist Unit
Wrist-view camera腕部视角相机
1920×1080 @ 30 Hz, 170° FoV, 62 g.1920×1080 @ 30 Hz，170° 视场角，62 g。
Ego Unit
Egocentric view第一视角
Stereo global-shutter RGB with IMU for task context.双目全局快门 RGB 加 IMU，提供任务语境。
Compute Unit
Main control主控
1 TB SSD, 2.5G Ethernet, 5× USB-C.1 TB SSD，2.5G 网口，5 个 Type-C。
From raw capture to training data从原始采集到训练数据
From gripper sessions to training-ready datasets从夹爪采集到可训练数据集
Capture采集
Wearable SenseXperience units record synchronized video, IMU, audio and interaction signals in real-world settings.SenseXperience 可穿戴模块在真实场景中同步记录视频、IMU、音频与交互信号。
Process后处理
IO-Agent-Lab custom-trained agents run timestamp alignment, pose estimation and gesture recognition.IO-Agent-Lab 定制训练的智能体完成时间对齐、位姿估计与手势识别。
Review & QA审核与质检
Automated and manual QA in EmbodiFlow filters noisy, incomplete or low-quality samples before training.EmbodiFlow 自动与人工质检流程，在训练前过滤噪声、不完整或低质量样本。
Export导出
Training-ready datasets in LeRobot, HDF5 and MCAP, compatible with ROS and popular robot-learning workflows.导出 LeRobot、HDF5、MCAP 等可训练格式，兼容 ROS 与主流机器人学习工作流。
FAQ
Frequently asked questions常见问题
What is UMI?什么是 UMI？
UMI (Universal Manipulation Interface) is an approach to collecting robot manipulation demonstrations with a handheld gripper and an attached camera, so people can gather data in real environments without a robot present.UMI（Universal Manipulation Interface，通用操作接口）是一种用带相机的手持夹爪采集机器人操作示范的方式，让人在没有机器人的真实环境中也能采集数据。
What does the SenseXperience UMI Unit record?SenseXperience UMI 单元记录哪些数据？
A 170° RGB camera (2048×1536 at 30 Hz), gripper jaw angle from a magnetic encoder at 240 Hz, IMU at 240 Hz and two-channel audio. It can be combined with wrist and egocentric units for additional views.170° RGB 相机（2048×1536 @ 30 Hz）、240 Hz 磁编码器测得的夹爪角度、240 Hz IMU 以及双声道音频；还可与腕部、第一视角单元组合，获得更多视角。
Which formats can I export?可以导出哪些格式？
Through EmbodiFlow, data is exported in formats such as LeRobot, HDF5 and MCAP, compatible with ROS and popular robot-learning workflows.通过 EmbodiFlow，数据可导出为 LeRobot、HDF5、MCAP 等格式，兼容 ROS 与主流机器人学习工作流。
Can grippers and sensors be customized?夹爪和传感器可以定制吗？
Yes. SenseXperience is modular: cameras, grippers, sensors and processing pipelines can all be customized, with options to add motion capture and tactile sensing.可以。SenseXperience 采用模块化设计，相机、夹爪、传感器与处理流程均可定制，并可加入动捕与触觉采集。
Start collecting manipulation data.开始采集操作数据。
Tell us about your tasks, robots and data requirements — we will help you configure the right capture system.告诉我们你的任务、机器人与数据需求，我们帮你配置合适的采集系统。