Module 3/5 · Weeks 7–9 · 27 h

Drones in warehouses

UAT 363 Unmanned Aircraft Systems Technology for Transportation and Smart Warehousing

About 90 minDraft, awaiting reviewLast updated 27 September 2026

By the end of this module you will be able to

Explain drone use in warehouses and the difference between stock taking and cycle counting
Plan rack-scanning flights and estimate the time required
Compare indoor positioning methods without GNSS, namely VIO, motion capture, UWB and AprilTag
Compute position from ranges to UWB anchors by least squares

Prerequisites: UAT 363 modules 1–2 · UAT 322 (navigation and SLAM)

Why this matters

A medical store holds thousands of items on high racks. Counting by hand needs forklifts, takes a long time and often stops work. Differences between real stock and records mean medicines run out or expire unnoticed. A drone scanning the racks can count more often, but indoors there is no GNSS signal, so the drone must know its position some other way.

Drones in the warehouse

The ETH Zurich report (Wawrla and colleagues, 2019) studied 12 real deployments in three groups: inventory management, intra-logistics, and inspection and surveillance. It concludes that inventory management has the highest potential, while intra-logistics is not yet feasible because of battery and payload limits. The main challenges are limited flying space, navigation without GPS, integration with existing systems, and battery safety.

The same report distinguishes stock taking, counting the whole warehouse usually once a year, from cycle counting, counting part of it daily or weekly. Drones suit cycle counting, because they can recount often without stopping work.

Top view of seven grey rack rows with six aisles between them. A pink line shows the drone flying up and down each aisle in turn — Figure 1. Stock-scanning path in the warehouse

Example 1. Number of locations and scan time

Assumed warehouse: 6 aisles, racks 40 m long divided into 16 bays and 5 levels high. The drone scans both faces of the aisle at once, level by level, at 0.5 m/s.

import math

AISLES, RACK_M, BAYS, LEVELS = 6, 40.0, 16, 5
SPEED, LEVEL_CHANGE_S, AISLE_CHANGE_S, BATTERY_MIN = 0.5, 10, 30, 15

locations = AISLES * 2 * BAYS * LEVELS
per_aisle_s = LEVELS * RACK_M / SPEED + LEVELS * LEVEL_CHANGE_S
total_s = AISLES * per_aisle_s + (AISLES - 1) * AISLE_CHANGE_S
flights = math.ceil(total_s / 60 / BATTERY_MIN)
print(f"{locations} locations, {per_aisle_s / 60:.1f} min per aisle, "
      f"total {total_s / 60:.1f} min -> {flights} battery flights")
print(f"daily cycle count of one aisle covers {2 * BAYS * LEVELS} locations "
      f"= all {locations} in {AISLES} working days")

960 locations, 7.5 min per aisle, total 47.5 min -> 4 battery flights
daily cycle count of one aisle covers 160 locations = all 960 in 6 working days

Counting the whole warehouse at once needs several battery changes, but cycle counting one aisle a day takes under 15 minutes and covers the whole warehouse every week.

Indoor positioning

Method	Principle	Strengths	Limits
VIO (visual-inertial odometry)	Onboard camera plus IMU estimates motion	No installation in the building	Drift accumulates; plain walls or low light are hard
Motion capture	Cameras around the room track markers on the drone	Very high accuracy	Expensive, limited area
UWB	Ultra-wideband radio ranging to anchors at known positions	Wider coverage, independent of light	Anchors must be installed; obstructions bias ranges
AprilTag	Visual fiducial markers at known positions	Cheap, clearly identifiable	Must be seen; used to correct drift

The PX4 guide explains feeding position from external systems (VIO or motion capture) into the EKF2 estimator and recommends a message rate of 30–50 Hz. UWB ranging is standardised in IEEE 802.15.4z-2020. The review by Alarifi and colleagues (2016) concludes that UWB performs better than other technologies for indoor positioning, while real accuracy depends on installation and environment and must be measured on site. AprilTag markers (Olson, 2011) serve both for positioning and as reference points on the racks.

A 20 by 10 metre room with four UWB anchors A1 to A4 at the corners. Blue dashed lines from each anchor to the drone are labelled r1 to r4 for the measured ranges — Figure 2. Positioning from ranges to UWB anchors

Example 2. Position from UWB ranges by least squares

The measured ranges carry small errors (assumed values). Subtracting the first equation from the others gives a linear system, solved by least squares.

import math

anchors = [(0, 0), (20, 0), (20, 10), (0, 10)]
true_pos = (7, 4)
noise = [0.05, -0.03, 0.04, -0.02]            # m
ranges = [math.dist(a, true_pos) + e for a, e in zip(anchors, noise)]

(x1, y1), r1 = anchors[0], ranges[0]
rows, rhs = [], []
for (xi, yi), ri in zip(anchors[1:], ranges[1:]):
    rows.append((2 * (xi - x1), 2 * (yi - y1)))
    rhs.append(r1 ** 2 - ri ** 2 + xi ** 2 - x1 ** 2 + yi ** 2 - y1 ** 2)

a11 = sum(r[0] * r[0] for r in rows); a12 = sum(r[0] * r[1] for r in rows)
a22 = sum(r[1] * r[1] for r in rows)
b1 = sum(r[0] * v for r, v in zip(rows, rhs)); b2 = sum(r[1] * v for r, v in zip(rows, rhs))
det = a11 * a22 - a12 * a12
x, y = (a22 * b1 - a12 * b2) / det, (a11 * b2 - a12 * b1) / det
print(f"ranges {[round(r, 2) for r in ranges]}")
print(f"estimate ({x:.2f}, {y:.2f}) m, error {math.dist((x, y), true_pos) * 100:.1f} cm")

ranges [8.11, 13.57, 14.36, 9.2]
estimate (7.01, 4.01) m, error 1.6 cm

Counting with images needs barcodes or labels on the boxes to be read. The RFly research (Ma and colleagues, 2017) instead used a drone as a relay to read and locate battery-free RFID tags in a warehouse, which does not require a direct view of the tag.

Module lab

Lab: scanning a mock rack

Set up a mock rack in the lab with a barcode or AprilTag on every slot
Apply Example 1 to the hospital store or lab dimensions and plan cycle counting
Install the lab’s indoor positioning system (UWB or motion capture), feed positions into PX4 following the guide, and measure the error against reference points
Apply Example 2 to real UWB ranges from several points and record the error
Fly a manual or automatic scan in an enclosed area and record the tag read success rate

Common mistakes

Watch out

Using GNSS-dependent flight modes indoors
Not surveying UWB anchor positions accurately, shifting every position
Flying near people and forklifts without separating time or space
Trusting VIO for too long without references to correct drift
Not checking the tag read success rate before concluding stock is missing

Summary

Inventory management is the best-suited warehouse task according to the ETH report, and suits cycle counting
Estimate scan time from aisles, levels and speed, then plan the counting cycle
Indoors, use VIO, motion capture, UWB or AprilTag to feed position to the flight controller
Ranges to several anchors give position by least squares

Check your understanding

Four aisles, two faces each, 20 bays per face and 4 levels. How many locations?
How many minutes does it take to scan 4 levels of a 30 m rack at 0.5 m/s?
How does stock taking differ from cycle counting?
Why does VIO need references such as AprilTags?
How many UWB anchors are needed at minimum for 2D positioning?

Answers

$4 \times 2 \times 20 \times 4 = 640$ locations
$4 \times 30/0.5 = 240$ s = 4 minutes
Stock taking counts the whole warehouse once a year; cycle counting counts part of it daily or weekly
VIO error accumulates over time, and references reset it
3 (not in a straight line); a fourth reduces the error

Key formulas

Range to an anchor	$r_{i} = (x - x_{i})^{2} + (y - y_{i})^{2}$
Linear equation after subtracting the first	$2 (x_{i} - x_{1}) x + 2 (y_{i} - y_{1}) y = r_{1}^{2} - r_{i}^{2} + x_{i}^{2} - x_{1}^{2} + y_{i}^{2} - y_{1}^{2}$

Key references

Wawrla, L., Maghazei, O., & Netland, T. (2019). Applications of drones in warehouse operations (White paper). ETH Zurich, D-MTEC, Chair of Production and Operations Management. link
PX4 Autopilot. Using vision or motion capture systems for position estimation. PX4 guide (main). link
Olson, E. (2011). AprilTag: A robust and flexible visual fiducial system. In 2011 IEEE International Conference on Robotics and Automation (pp. 3400–3407). IEEE. link
IEEE. (2020). IEEE standard for low-rate wireless networks — Amendment 1: Enhanced ultra wideband (UWB) physical layers (PHYs) and associated ranging techniques (IEEE Std 802.15.4z-2020). link
Alarifi, A., Al-Salman, A., Alsaleh, M., Alnafessah, A., Al-Hadhrami, S., Al-Ammar, M. A., & Al-Khalifa, H. S. (2016). Ultra wideband indoor positioning technologies: Analysis and recent advances. Sensors, 16(5), 707. link
Ma, Y., Selby, N., & Adib, F. (2017). Drone relays for battery-free networks. In Proceedings of ACM SIGCOMM 2017 (pp. 335–347). ACM. link

In class / field

Intensive lab and field practice recorded in a lab notebook

Learning evidence: Lab notebook signed by the instructor

Module quiz

This is a formative self-check, not a graded exam

Knowledge domain: Delivery, indoor operations and warehousing · Control, autopilot and navigation

Drones in warehouses

By the end of this module you will be able to

Why this matters

Drones in the warehouse

Indoor positioning

Module lab

Common mistakes

Summary

Check your understanding

Key formulas

Key references

Further reading

Drones for warehouses

Indoor positioning systems

In class / field

Module quiz

Drones in warehouses

Lesson

By the end of this module you will be able to

Why this matters

Drones in the warehouse

Indoor positioning

Module lab

Common mistakes

Summary

Check your understanding

Key formulas

Key references

Further reading

Drones for warehouses

Indoor positioning systems

In class / field

Module quiz