Physical AI data annotation
for robots and vehicles.
Label cameras, LiDAR, radar and telemetry in sync, in one platform, with AI that pre-labels and tracks objects.
What is Physical AI data?
Physical AI is models that perceive and act in the physical world: manipulators, mobile robots, drones and autonomous vehicles. Physical AI data is what they learn from: video, images, LiDAR and radar recorded together, and annotated in sync across every sensor.
Video
Footage from one camera or a whole rig, played as a synchronized group, with AI tracking through every frame.
Video labelingMulti-view images
Images of one moment from several cameras or angles, side by side, with the same object linked across views.
Image labelingLiDAR and RADAR
Point clouds as single scans or episodes, with the camera images of each frame and their calibration.
3D point cloudsRobotics data annotation.
Teams building manipulators, mobile robots and drones label their robot data in Supervisely: camera video, multi-view images and LiDAR, in sync. When a recording has more streams, our engineers build the viewer, the checks and the export it needs.
Multi-camera episodes
Every camera of a recorded episode plays in sync beside a 3D view of the scene, with a timeline of each tracked object.
Hand demonstrations
Head-camera video with hand keypoints next to a 3D scene built from depth, and the demonstrator’s hands retargeted to robot hands.
Richer training data
Objects, masks, keypoints and 3D cuboids turn robot demonstrations into training data a policy can learn more from.
Built around your recordings
When your data has streams no toolbox covers yet, the engineers behind the platform build the tools for it with Supervisely Apps.
Sensor fusion: cameras, LiDAR and radar in one scene.
Point clouds open with the camera images of the same moment and their calibration, and videos of one scene play as a synchronized group. Each label is checked twice: in 3D and on the picture.
From 2D labels to 3D objects
Outline an object on a camera image with a box, polygon, brush or the Smart Tool, and the matching 3D object appears in the cloud.
Multi-camera video in sync
Videos from different cameras play together, and an object seen by several cameras stays one object.
Tracked from scan to scan
A cuboid placed once is tracked through the episode, and interpolation fills the frames between two keyframes. RADAR clouds get the same tools as LiDAR.
AI-assisted labeling at scale.
Annotators review labels instead of drawing them: in point clouds, on camera images and in video.
A 3D AI assistant with no training
Class-agnostic: one click fits a cuboid, the 3D Smart Tool places one around a circled object, and ground segmentation separates the ground from everything on it.
Your own 3D detectors
Serve MMDetection3D detectors such as CenterPoint, PV-RCNN and PointPillars to pre-label whole projects, or train them on your point clouds.
Tracking in video and episodes
AI tracking carries each object through the footage, and cuboids are tracked from scan to scan; annotators fix only the frames where a track slips.
The data loop: from recordings to trained models.
Hard episodes go back to labeling, models retrain on the result and pre-label the next batch, in one platform.
- 01Data curationBring any format or bucket into one place, catch label errors before training and version each state of a dataset a model learns from.
- 02Labeling workflowHand out tasks to your own people and outside vendors, check their work with consensus and review, and measure every annotator.
- 03Model trainingTrain models on your labeled data without code, then deploy them on your own GPUs to pre-label new data.
Enterprise: on-premise and private.
Robot and vehicle recordings often cannot leave the company. The platform used by 120,000+ engineers in its community also runs where your data lives.
Your servers or private cloud
Run Supervisely on your servers, in a private cloud or in a Kubernetes cluster, or fully offline in a network without internet access.
No trackers
The self-hosted Enterprise Edition has no trackers, does not need an internet connection and can work inside your private network.
Single sign-on and roles
Only approved people get access to your data, with single sign-on through LDAP or OpenID Connect and custom roles.
Why Supervisely: computer vision experts since 2013.
Supervisely began in 2013 as a computer vision consultancy that built end-to-end solutions for clients from the US to Japan, and has built its own platform since 2017. The engineers behind it work with your team from the first day of the trial.
- 2013Started as a computer vision consultancy
- 2017Supervisely released in August
- 120,000+Researchers and engineers worldwide use Supervisely
- Fortune 500Companies trust Supervisely
Engineers who learn your case
Specialists who understand your data and your task suggest a setup for it, and an engineer stays at your side through installation and a guided trial of 30 days or more.
A dedicated success manager
A private Slack workspace carries your team through onboarding, and training sessions come from the people who build the platform.
Built around your pipeline
When a feature is missing, we help build it with Supervisely Apps and custom integrations.
Proven on sensor fusion
Renault Group teams work with camera and LiDAR data across sensor fusion projects in Supervisely.
Here’s why teams trust us.
“A very high product quality.”
BMW Group uses Supervisely to annotate manufacturing images from production lines in their world-wide plants for enhancing quality inspections using deep learning. Supervisely is integrated into the BMW Group AI Platform in order to empower computer vision based AI use cases.
BMW Group · Manufacturing quality inspection“Immensely grateful for their pivotal role in our work”
Working with Supervisely has significantly enhanced our capability to develop AI models for lung CT scans. What sets Supervisely apart is its exceptional support team who are really responsive and adapt to our unique needs with a range of apps and helper files.

“Helps us to do data scientist work better and faster”
We use Supervisely since 2019. The key advantage of this tool is that Supervisely provides a complete data treatment pipeline. An important advantage is that a Supervisely instance can be deployed autonomously on a Client infrastructure, and distributed on different servers.

“A valuable part of our computer vision operations”
Supervisely is a valuable part of our computer vision operations, enabling our teams to efficiently work with camera and LiDAR data across sensor fusion projects. We use Supervisely for its extensive feature set and its ability to adapt to our changing requirements.
“Extremely satisfied with Supervisely and would highly recommend it”
The review tools have been invaluable in ensuring the quality and accuracy. The Python SDK has also been incredibly helpful in automating and streamlining our workflow. The support team on Slack has been extremely helpful and responsive.

“Supervisely excels at that and much more”
We’ve been exceedingly impressed with the customer support, addition of new features, and the flexibility of the publicly available SDK/API. The Supervisely team has also been fast to respond to support questions, and has shown a lot of openness when given feedback on potential improvements.

Your team next?
Talk to a computer vision engineer about your data, your models and your infrastructure.
Book a demoFrequently asked questions.
Which annotation platforms support sensor fusion labeling for synchronized LiDAR, radar and camera data?
Supervisely labels LiDAR and RADAR point clouds, as single scans or episodes, together with the camera images of each frame and their calibration. Each 3D object is projected onto the images, and a 2D box or mask drawn on an image becomes the same object in 3D. See the 3D point cloud annotation tool.
What tools handle labeling across video, LiDAR and sensor-fused datasets for robotics teams?
Supervisely is one platform with a labeling toolbox for each data type: images, including multi-view groups; video, with synchronized multi-camera views and GPS telemetry; and LiDAR and RADAR point clouds and episodes with their cameras. See the platform overview.
Can Supervisely annotate robot demonstration data?
Yes. Robot camera video, multi-view images and LiDAR are labeled in sync, with objects, masks, keypoints and 3D cuboids. For streams no toolbox covers yet, our engineers build a viewer and checks for your recordings with Supervisely Apps.
Can I pre-label mixed point cloud and camera data?
Yes. The class-agnostic 3D AI assistant fits cuboids and segments the ground without training on your data, and you can serve MMDetection3D detectors such as CenterPoint, PV-RCNN and PointPillars to pre-label whole projects. On the camera images, boxes, polygons, masks and the Smart Tool work as well. See AI-assisted labeling.
Which autonomous driving formats can I import and export?
Auto Import reads PCD, PLY, LAS and LAZ files and annotated datasets in KITTI 3D, KITTI-360, SemanticKITTI, nuScenes, Lyft, ROS bag and Supervisely format. Export writes Supervisely JSON, KITTI 3D, nuScenes, SemanticKITTI and ROS bags. See the supported point cloud formats.
Can Supervisely run on-premise for sensitive robot or vehicle data?
Yes, with the Enterprise Edition. It runs on your own servers, in a private cloud or in a Kubernetes cluster, or fully offline in a network without internet access, with single sign-on and custom roles. See security.
What is Physical AI training data?
It is the data that models acting in the physical world, such as manipulators, mobile robots, drones and autonomous vehicles, learn from: video, images, LiDAR and RADAR point clouds recorded together, and annotated in sync across every sensor.