For most Indian warehouses and godowns with selective racking a few levels high, a disciplined manual cycle count or a camera rig pushed along the aisle is the simplest way to run rack audits. Drones and autonomous scanning robots make more sense in very large, tall, high-bay sites, where counting the top levels by hand is slow and risky and the site is big enough to keep a fleet busy.
This guide compares five approaches: manual physical verification, a camera rig moved along the aisle, forklift-mounted cameras, drones and autonomous scanning robots. It looks at what each one counts, the space it needs, safety and permissions, people, disruption, and the evidence it leaves behind.
What a rack audit has to prove
A rack audit checks that what is physically in each bin matches what the system says. In India this is not only an operations question. Under CARO 2020, the statutory auditor reports on whether the coverage and procedure of management’s physical verification of inventory is appropriate, and on discrepancies of 10% or more in aggregate for each class of inventory (BCAJ on CARO 2020). So whatever method you choose, it has to leave a record an auditor can follow.
There are two ways to organise the counting. A full wall-to-wall count checks everything in one event and usually needs operations to pause. Cycle counting checks small portions of stock frequently through the year and runs alongside normal work (ERP Research glossary). All five methods below can support cycle counting; they differ in how much of the rack they cover per hour, how high they reach, and what they need from your site.
The five options
1. Manual count with scanners and tally sheets
Counters walk the aisle with a handheld scanner or a printed tally sheet, read the bin label, count boxes and write down the number. For upper levels they need a reach truck, order picker or ladder. It needs no new equipment and works in any layout. The costs are people, time, and the risk of miscounts and transcription errors. Evidence is usually the signed sheet, not a photo.
2. Camera rig moved along the aisle
A trolley or mast carrying cameras is pushed or driven down the aisle. Software reads each bin label from the images and counts the visible boxes, and a person reviews the results. It needs one or two operators instead of a counting team, and it leaves an image for every bin. Its reach is limited by the height of the rig, so check that it covers your top beam level.
3. Forklift-mounted cameras
Cameras ride on the trucks you already use. Gather AI says its MHE Vision “mounts on existing forklifts, reach trucks, and order pickers” and captures inventory continuously across the shift, with integration to the WMS (Gather AI MHE Vision). The appeal is that counting happens as a by-product of normal work. Because the cameras go where the trucks go, ask the vendor how locations that rarely see a truck get covered.
4. Drones
Indoor drones fly the aisles and photograph or scan the rack faces. Corvus Robotics says its drones navigate without GPS using cameras and neural networks, need no stickers, reflectors or beacons, scan barcodes, and can work 24/7 with warehouse lights on or off, charging at docks placed at the ends of racks (MIT News). Its site says it works with any WMS via API or standalone and that it is supported across the US (Corvus Robotics). Gather AI’s Drone Vision flies autonomously, is aimed at “high-bay and large-footprint racking”, and reports 3D case counts, empty locations and LPN-to-location checks (Gather AI Drone Vision).
5. Autonomous scanning robots
A wheeled robot with a tall sensor tower drives the aisles on its own. Dexory describes a tower that stands at a minimum of 3 m and extends to 12 m, with a top camera that measures racks up to 14 m, and a battery that gives about 4 to 5 hours of scanning (Dexory). Dexory’s site says the robot covers full-height racking in one pass, captures point cloud and LiDAR data, runs alongside existing work, and needs a clear 3 m x 3 m charging dock area and flat, dry floors (Dexory home page).
Side-by-side comparison
| Factor | Manual count | Camera rig in the aisle | Forklift cameras | Drones | Scanning robot |
|---|---|---|---|---|---|
| What it counts | Labels and boxes, by eye | Bin labels and visible boxes | Depends on vendor; Gather AI lists case counts and LPN checks | Barcodes/labels; some report case counts and empties | Full rack face; vendor-dependent detail |
| Height reach | Limited by ladder or truck | Limited by rig height | As high as the forks travel | Suited to high-bay | Dexory: racks up to 14 m |
| Aisle and floor needs | Any | Room to push the rig | None beyond normal truck use | Clear airspace; docks at rack ends | Flat, dry floor; charging area |
| People needed | Counting team | One or two operators plus a reviewer | Existing drivers | Supervisor and exception review | Supervisor and exception review |
| Disruption | High for wall-to-wall counts | One aisle at a time | Low; rides on normal work | Vendor-dependent; see safety | Vendors say it runs alongside work |
| Evidence | Signed sheet | Image per bin | Captured observations, vendor-dependent | Scan record per location; images vary by vendor | Images and 3D scan data |
Factors that decide it
What gets counted: labels or boxes
Reading a bin or pallet label tells you what should be there. Counting boxes tells you how much is there. Many audits need both. Ask every vendor whether the system reads the location label, the product label, or both, and how it counts boxes stacked behind the front row. Gather AI, for example, lists a “Smart Multi-Deep Logic” feature (Gather AI). Systems that only see the front face will miss stock behind it.
Aisle width and ceiling height
Height is the biggest single factor. If your top beam is within easy reach of a person on a reach truck, manual counts and aisle rigs stay practical. As racking gets taller, the time and risk of counting the top levels by hand rises, and tall robots or drones start to pay their way. Narrow aisles limit what you can push or drive through, so measure the clear width between rack uprights and any pallet overhang, not just the drawing.
Lighting
Cameras need to see the label. Upper levels in many godowns are dimmer than the floor, and labels may face away or sit under shrink wrap. A rig or robot can carry its own lights; ask whether it does. Corvus says its drones can work with warehouse lights on or off (MIT News).
Safety and permissions
Anything moving in a live aisle needs a safety plan. Vendors disagree on drones: Corvus says its drones operate safely around people and forklifts (MIT News), while Dexory, which makes ground robots, writes that aisles need to be closed during drone scanning (Dexory). Ask to see it working in a site like yours.
In India, drones are also regulated. The Drone Rules, 2021 require registration on the Digital Sky platform for a unique identification number and a remote pilot licence for most operation, with some exemptions for nano and non-commercial micro drones (iPleaders overview; Majmudar & Partners). The summaries we reviewed do not spell out indoor warehouse flights, so confirm the position with your legal team before you plan a drone pilot.
People and disruption
Manual counts are labour-heavy, and full counts often stop picking. Forklift cameras add almost no work. Drones and robots run on their own but still need someone to clear exceptions. An aisle rig sits in between: a small crew, one aisle closed for a short time.
Evidence and reconciliation
The count is only useful once it is reconciled with the WMS. Check what the system outputs (a CSV, a report, or a live WMS feed), who approves a mismatch, and whether each bin keeps an image. A photo per bin turns an argument about a count into a quick look.
Where drones and robots fit better
Be fair to the autonomous options. In a very large distribution centre with tall racking, a robot or drone fleet can count far more locations, more often, than any walking team, and it can work overnight. Corvus says a 1-million-square-foot facility takes about a week to be fully operational (MIT News), and Dexory says such a warehouse is typically finished in under one shift (Dexory home page). If that describes your site, shortlist them. If your site is smaller, lower, or has narrow and crowded aisles, a manual count or an aisle rig is usually easier to start with. Also check whether each vendor supports sites in India.
How DeepVision Warehouse fits
DeepVision Warehouse from Indus Vision, Bengaluru, takes the camera-rig approach. A rig with front and top cameras on both sides is moved along the aisle; it reads each bin label and counts boxes, the auditor approves or rejects each bin, and the reconciliation comes out as CSV. It runs on site and needs no WMS change: Excel or CSV in, XLSX, CSV or PDF out. It was built in an FMCG distribution centre. The same product also covers dock tally at the loading conveyor. For the terms used here, see our glossary of warehouse counting terms.
Frequently asked questions
Is cycle counting enough for statutory physical verification in India?
It can be, if the plan covers all material items, typically at least once a year or under a documented systematic plan. Under CARO 2020 the auditor comments on whether the coverage and procedure are appropriate (BCAJ). Agree the approach with your auditor.
Can a drone count boxes, or only read labels?
It depends on the vendor. Corvus describes barcode scanning, while Gather AI lists 3D case counts and empty detection for its drone product. Ask for a demo on your own SKUs.
Do I need to change my WMS to use camera-based audits?
Not always. Some vendors integrate through an API; others, including DeepVision Warehouse, work from Excel or CSV files. Check what each one needs before you start.
Which option disrupts operations least?
Forklift-mounted cameras ride on normal work, and scanning robots are designed to run alongside it. An aisle rig closes one aisle for a short time. Wall-to-wall manual counts usually disrupt the most.
How should I run a trial?
Pick a few aisles with a mix of high and low levels, fast and slow movers, and awkward labels. Count them manually first, then compare each system’s result and the time it took, bin by bin.