Back to selected work
Case 11 · Hangar · Warehouse Inventory Fleet · Shipped 2024
// Case 11 · 2024 · Indoor Autonomy · Warehouse

Hangarwarehouse fleet.

A fleet of indoor-autonomous inventory drones for a 220k-sqft distribution centre. Visual-SLAM in GPS-denied aisles, barcode + OCR at fly-by speed, cycle counts in three hours.

EngagementFixed-price
StackROS 2 · SLAM
Timeline16 weeks
StatusIn production
hangar.internal/fleet
Hangar — warehouse drone fleet dashboard
Chapter 01 · The Brief

Cycle count took
a full weekend. Every time.

The distribution centre ran cycle counts every quarter — closing the facility for a full weekend, pulling staff off regular operations, and scanning 28,000 pallet locations by hand. Errors crept in; discrepancies between the WMS and physical inventory cost the client $120k in write-offs the year before we started.

The brief: a fleet of autonomous drones that flies the aisles at night, reads barcodes and OCR labels at fly-by speed, and reconciles against the WMS without shutting the facility or touching a scanner.

Brief at a glance
Client
3PL distribution centre, 220k sqft
Inventory
28,000 pallet locations · 4 aisle blocks
Problem
Quarterly cycle counts: full weekend shutdown · $120k write-offs
Goal
Nightly incremental counts · zero facility downtime
Constraint
GPS-denied environment · 8m rack height · forklift traffic
Chapter 02 · The Approach

Navigate by walls,
read every label.

Indoor Navigation

GPS is useless inside a steel-rack warehouse. Each drone builds and maintains a Visual-SLAM map using a downward stereo camera and rack-mounted AprilTag anchors for loop-closure corrections. A ROS 2 navigation stack handles path planning, obstacle avoidance for forklifts and personnel, and aisle-by-aisle task assignment across the fleet.

Drones operate at night between shifts, dock autonomously on charging pads between aisles, and hand off tasks when battery drops below threshold — no human intervention required after the evening launch.

Barcode & OCR at Speed

A high-speed global-shutter camera with a dedicated ring-light flash captures barcode and label images at fly-by speed — up to 1.8 m/s past a rack face. An on-board NVIDIA Jetson runs a custom YOLO detector to locate label regions, feeds them into Tesseract OCR with a WMS-vocabulary language model, and achieves 98% read accuracy without slowing down.

Results sync to the WMS in real-time over Wi-Fi. Discrepancies are flagged with location and confidence score for human review the next morning.

Chapter 04 · By the numbers
1.4k
Pallets scanned
per hour.
per drone
98%
Barcode read
accuracy.
in production
16×
Faster than
manual count.
3h vs full weekend
0
Facility
shutdowns.
runs overnight
Chapter 05 · Inside the fleet

Launch it at night,
read it by morning.

Fleet dashboard, SLAM map, live scan feed, WMS reconciliation, and discrepancy report — the full warehouse pipeline.

// 01 · Fleet
Hangar — fleet status dashboard

All drones, battery level, current aisle, and task queue.

// 02 · Map
Hangar — Visual-SLAM warehouse map

Live SLAM map with drone positions and completed aisles.

// 03 · Scan
Hangar — live barcode scan feed

Real-time barcode and OCR reads syncing to the WMS.

// 04 · Reconcile
Hangar — WMS reconciliation view

Scanned inventory vs WMS — discrepancies flagged instantly.

// 05 · Report
Hangar — morning discrepancy report

Location, label, and confidence score for every mismatch.

click to expand · drag to explore
Closing

The
credits.

  • Engagement
    Fixed-price · 16-week program
  • Airframe
    Custom quad · 8 units · auto-docking charging pads
  • Navigation
    ROS 2 · Visual-SLAM · AprilTag anchors · Nav2
  • Comms
    MAVROS · Wi-Fi mesh · WMS REST API
  • Perception
    NVIDIA Jetson · YOLO label detection · Tesseract OCR
  • Dashboard
    Next.js · WebSocket fleet telemetry · live SLAM map viewer
  • WMS Integration
    REST sync · nightly reconciliation report · email digest
  • Status
    In production · nightly operation · 1 DC client