Every year, $1.1 trillion in goods disappears into what the logistics industry calls inventory distortion—the gap between what a warehouse management system says is on a shelf and what's actually there. Even in well-run distribution centers, manual cycle counting achieves only 63–80% accuracy. A full physical inventory of a large DC takes weeks, happens twice a year, and exposes workers to ladder and forklift hazards throughout. The operational premise of autonomous warehouse inventory drones is disarmingly simple: fly a robot through the aisles every night instead.

Why GPS Is the Wrong Starting Point

Satellite signals don't penetrate metal-racked distribution centers. Building envelopes and rack infrastructure block GPS completely, which means the standard positioning stack for outdoor UAS is dead on arrival the moment a drone crosses the loading dock. The commercial systems that have reached scale solve this through sensor fusion: visual-inertial odometry (tracking position by analyzing changes in camera frames between successive moments), LiDAR for spatial mapping, and SLAM—Simultaneous Localization and Mapping—which lets a drone build a map of its environment while simultaneously locating itself within it, with no pre-installed beacons or markers required.

In practice, the leading vendors have pushed past the academic baseline. Corvus Robotics' Corvus One carries 14 cameras and navigates using neural networks rather than pre-programmed paths; a 1-million-square-foot facility can be commissioned in approximately one week, with only dock installations at rack ends required. Mohammed Kabir, CTO of Corvus and a 2021 MIT graduate, described the system's lineage directly: “We were the first in the world to ship a learning-based autonomy stack for an indoor aerial robot using machine learning.” Gather AI's platform—updated to the US-made ModalAI Starling 2 Logis drone available Q2 2025—adds dynamic obstacle avoidance and depth sensing via the VOXL 2 autopilot, requiring no GPS, Wi-Fi infrastructure changes, or floor markers. Verity, the Zurich-based firm whose CEO Raffaello D'Andrea co-founded Kiva Systems before its 2012 Amazon acquisition, runs its drones in complete darkness without operators present, using a custom indoor positioning system that guides flight above storage level three and handles obstacle detection with automatic path rerouting.

What the Drones Actually Scan—and What Happens Afterward

The baseline is 1D/2D barcode and QR code reading on pallet labels. The current generation goes considerably further. Gather AI's computer vision reads lot codes, freeform text, expiration dates, case counts per pallet, and slot occupancy—inventory intelligence that doesn't require a discrete scannable label at all. Its Starling 2 drones process 1,500 pallets per hour (25 times faster than manual) in aisles as narrow as 4 feet 5 inches, reaching up to 99.9% inventory accuracy while reducing manual counting labor by 80%. Dexory's platform adds LiDAR-derived 3D spatial data, addressing the angle and depth constraints that aerial drones face in double-deep racking configurations.

The data pipeline out the back end matters as much as the scan itself. Drone reads stream timestamped with pose data to an edge gateway over Wi-Fi, pass through GS1 validation and de-duplication, then post to the warehouse management system via REST, GraphQL, or MQTT. The WMS remains the system of record; the drone is the sensor. Idempotent posting—preventing double-counting on communication retries—and complete audit trails linking flight missions to inventory documents are baseline requirements for enterprise integration. Mismatches surface as exceptions in near-real-time, improving replenishment decision latency substantially over weekly manual exception reports. Deployments that integrate cleanly typically expose 1–2% phantom stock within weeks and sustain greater than 99% location accuracy on an ongoing basis.

Four Vendors, Four Approaches

Verity operates in dozens of warehouses globally. Its IKEA partnership, begun in 2021 at a Swiss distribution center, had grown to 250+ drones across 73 locations in 9 countries by 2024. Distribution centers run drones continuously; retail stores restrict flight to off-hours when customers are not present in the warehouse. UPS Supply Chain Solutions added Verity in September 2024, joining On (the Swiss running shoe brand) on the customer roster.

“The completion of our Series B funding round is an important nod to the value our system provides to clients.”
— Raffaello D'Andrea, Founder and CEO, Verity

Corvus Robotics' disclosed customer outcomes are the most granular in the record. Southern Glazer's Wine & Spirits scaled across 9 distribution centers, achieving 6x more frequent cycle counting and locating 35,000+ verified discrepancies. Dermalogica increased counts from monthly to 52 times annually, reallocated 120 labor hours per month, and runs lights-out with no on-site operators—exactly the kind of unattended operation that manual counting can never reach. MSI Surfaces achieved 20x faster cycle counting, reallocated $110,000 in labor per site, and recovered $1.25 million in previously lost inventory. In February 2026 Corvus added a Cold Chain variant rated to −20°F, extending autonomous inventory to freezer warehouses. The company also offers Corvus Trident, an AI copilot mounted on forklifts for real-time pallet tracking as a complementary data layer.

Gather AI, founded by Carnegie Mellon alumni in Pittsburgh in 2017, counts GEODIS, AAFES, Kwik Trip, Barrett Distribution, and dnata among its named customers. CEO Sankalp Arora positions the company's software-intelligence emphasis as deliberate architecture: “Our hardware-agnostic AI and computer vision stack empowers logistics customers with unparalleled flexibility and access to cutting-edge solutions.”

Dexory—UK-based—uses wheeled ground robots rather than aerial drones but solves the identical problem with complementary strengths: 10,000 locations per hour, better handling of double-deep and block-stacked configurations, and 99.9% real-time accuracy. Maersk achieved 99% inventory accuracy, a 50x reduction in WMS errors, and saved 10 hours daily on stock corrections within one week of deployment. Yusen Logistics cut monthly counting time from 100 hours to 2 hours with zero lost items. Iron Mountain increased accuracy from 97.8% to 99.9%. Dexory's February 2026 next-generation robot extended scanning range from 40 to 60 feet via a modular architecture; its “Storage Health” AI feature now flags damaged racking, defective pallets, unstable loads, hanging shrink wrap, and crush damage alongside count data.

The Economics, and Where the Walls Are

The business case for a mid-size distribution center is direct: $200,000–$500,000 in annual labor savings, with payback on initial deployment in 6–12 months. Gather AI reports under-six-month ROI for its deployments. GNC Holdings provides a representative operational picture—cycle counts increased from 2–4 per year to 10–12, while inventory control staff dropped from 20 to 13. As Bill M., VP of Distribution at GNC, put it: “The drones are doing things with limited individual value, but deliver immense business value.”

The constraints are equally concrete. Minimum aisle width is a hard gate: Gather AI's Starling 2 requires at least 4 feet 5 inches, excluding some high-density configurations. Wi-Fi reliability is a prerequisite—patchy coverage breaks the data pipeline at the edge gateway. Battery consumption increases with altitude, limiting practical ceiling heights. And deployment requires a WMS capable of ingesting drone data, which frequently means infrastructure work before the first flight. Dexory's position that aerial drones face inherent angle and accuracy limitations in double-deep racking represents a genuine constraint, not marketing copy—the physics of camera angle at height in tight bays is real.

Market consolidation is underway. Ware, an early aerial inventory startup, was acquired by Stow Group, a rack manufacturer—a pattern consistent with a market moving from venture-funded pilots toward embedded industrial product lines. The global warehouse drone market was projected at $7.2 billion by 2030 at 12.2% CAGR, as of a January 2026 analysis. With Verity operating across dozens of active warehouse deployments, Corvus serving Fortune 500 companies across multiple verticals, and Dexory operating for Maersk, DHL, GXO, and DB Schenker, the technology has moved decisively past proof-of-concept. Over 150,000 warehouses worldwide still use manual inventory checks. The limiting factor for the next adoption wave is integration readiness, not whether autonomous indoor scanning works.

Sources