Contact Us
450 TASK HOURS · OUTDOOR RIDER

First-person delivery rides captured on chest / helmet cams — pickup, ride, hand-off, edge-case yields. NYC, NJ, and Boston-area routes. 60-second sample shown here.

L1 + L3 + L5 ANNOTATION

2D detection + tracking (23 IDs across 36 frames) · ego action / state · bilingual captions at clip + event + keyframe granularity — packaged in an OpenLabel-compatible JSON.

EVENT-DENSE SAMPLING

0.5 fps base + extra keyframes at every risk moment, mount/dismount, package handoff and crosswalk-yield event so the rare-but-critical frames never get diluted.

RISK-FLAGGED FOR WORLD MODELS

Every clip carries a T1/T2/T3 risk timeline (close-pass, hard-brake, run-red, salmon, etc.) with named causes — drop-in ready for VLA pre-training and safety-prior models.

SCENE COVERAGE · OUTDOOR DELIVERY
450 task hours
across NYC / NJ / Boston · 6 weather buckets · 60-second sample below
240 HOURS
Risk events covered
Hard brake · Close-pass <1m · Mid-block ped crossing · Crosswalk yield · Wrong-way (salmon) · Sidewalk riding · Run-red / Run-yellow · Phone glance · Door-zone hazard · Cone / barrier weave · Dog leash crossing · Truck blind-spot pass
DELIVERY
SCENARIO
EGO POV · CHEST CAM
MODE walking / mounted / riding
SPEED stopped → fast
BRAKE none → emergency
EVENTS L4 risk / action
SHIPPED WITH CLIP RGB × 1 (ego POV, 60 s · 30 fps) L1 bbox × 114 · 23 track IDs L3 ego state per annotated frame L4 risk events × 3 (T2 + 2× T3) L5 captions × 55 (clip + event + keyframe) Bilingual (en + zh) OpenLabel-compatible JSON
00:00 / 01:00
CLIP
SOURCE
SCENE
INIT