LIDAR 3D Point-Cloud Annotator

PROSYS LTD builds high-quality training data for frontier AI — from agentic benchmarks used by frontier labs to 3D perception data for autonomous driving. We recently kicked off a LIDAR annotation engagement for an autonomous trucking program and are growing a dedicated annotation cell for it.

You will annotate objects in LIDAR point-cloud scenes captured from a vehicle-mounted sensor rig. Working in the IMU coordinate frame, you will place oriented 3D bounding boxes (position, size, yaw) around vehicles, pedestrians, cyclists, traffic cones and other road objects, assign the correct class from a 24-class taxonomy, and keep object IDs consistent across frames. Every delivery passes acceptance QA: box fit is checked against ground truth at 3D IoU ≥ 0.7, headings must match the direction of travel, rigid objects must keep constant size across frames, and ghost boxes or missed objects fail the batch.

• Draw and fit oriented 3D bounding boxes on point-cloud objects with more than 10 points (cones: 5+), tightly enclosing the object without including motion-blur trails, mirrors or open doors
• Classify every box correctly across the 24-class taxonomy (Car, Truck, K-truck handling, Tractor/Trailer articulation, Motorcyclist vs Motorcycle, groups, and more)
• Set accurate yaw/heading for every object and maintain consistent object IDs across consecutive frames, including short occlusions
• Flag unusable frames (misaligned merges, missing data) with reasons instead of annotating them
• Hit and hold the acceptance bar: 3D IoU ≥ 0.7, correct classes and headings, no floating/sunken boxes, no ghosts, no size jumps
• Work inside our QA loop: respond to reviewer feedback and re-annotate rejected batches quickly

• Strong 3D spatial reasoning — you can read a sparse point cloud and see the vehicle in it
• Obsessive attention to detail and the patience for repetitive, precision work
• Comfortable desktop setup: mouse (not trackpad), stable internet, and a machine that can run a WebGL 3D editor
• Written English good enough to follow a detailed technical specification
• Availability of at least 20 hours per week with consistent daily output

• Prior LIDAR / point-cloud annotation experience (Supervisely, SUSTech POINTS, Xtreme1, CVAT or similar)
• Background in autonomous driving, robotics, GIS, 3D games or CAD
• Experience working against formal QA acceptance criteria (IoU thresholds, defect ratios)

1. 1Apply with the short form on this page (about 5 minutes)
2. 2Pass the guideline quiz — multiple choice on the annotation specification
3. 3Complete the hands-on 3D annotation assessment in your browser (scored automatically against ground truth)
4. 4Interview + paid calibration batch, then production onboarding