---
title: "The Forklift Is One of the Last Machines in the Warehouse Without a Driver — So Far"
date: 2026-11-16
category: Humanoids & Robotics
site: NeuroAI
canonical: https://neuroai.site/a/na-robot-autonomous-forklift
language: en
---

# The Forklift Is One of the Last Machines in the Warehouse Without a Driver — So Far

> Chinese driverless forklifts now sell in Japan and Europe. The hard part is not the driving — it is the 2.1-metre aisle and the pallet nobody stacked straight.

A 1.5-tonne cage of car parts has to go four layers up, in a building where people are still driving manual forklifts two aisles over. The machine doing it has no cab.

Warehouses have been automating for twenty years. Conveyors, shuttle systems, shelf-carrying AMRs — all standard. The forklift survived because it is the one machine that has to cope with whatever the previous human left behind: a pallet stacked off-centre, a damaged cage, a trailer loaded by hand.

That is the problem a group of Chinese companies has been attacking, and this year they are doing it in the markets that build forklifts.

## What "autonomous" means here

An unmanned forklift (无人叉车) is not a new vehicle class. It is a standard counterbalance or reach truck chassis with the driver replaced by a perception stack, a controller and a fleet system.

VisionNav Robotics, founded in 2016, is one of the most visible Chinese suppliers. Its catalogue spans seven series — counterbalance stackers, counterbalance trucks, reach trucks, slim forklifts for very narrow aisles (超窄巷道), pallet jacks, tow tractors and AMRs. The company says it has deployed more than 750 projects across 30-plus countries, with sales operations in the US, Europe and Japan.

The published case studies are specific enough to be checkable, which is unusual in this sector:

- A **VNP15** counterbalance stacker rated at 1,500 kg with a 2,400 mm lift, doing four-layer stacking of finished-goods cages at an automotive parts plant. Three cage sizes, fork pockets narrow enough that manual drivers were mis-stacking. Positioning accuracy quoted at ±10 mm, detection range 600–2,600 mm, and a vision check that approves or rejects each placement before the load is released.

- A fleet of ten **VNA15** units in a logistics warehouse with aisle spacing as tight as about 7 feet (2.1 metres), integrated with the customer's WMS and WCS.

- In Japan, a local subsidiary sells a slim series specified for 2.2-metre aisles, marketed with the phrase that gives away the whole sales pitch: no facility modification.

The 2.1-metre and 2.2-metre figures come from different product lines, not from two measurements of the same truck.

That last point is the commercial core. The navigation is 3D SLAM (激光SLAM) with laser and vision, so the truck maps the building itself. No magnetic tape in the floor, no reflectors on the racking, no shutting the aisle for a week to install infrastructure. For an operator with a 1990s building and a two-year payback threshold, "we can deploy without touching your floor" is worth more than any spec on the datasheet.

## Why this is harder than the AMR wave

The shelf-carrying robots that took over e-commerce warehouses work in a controlled grid: known aisles, known totes, no people. A forklift works in the opposite conditions.

- **Mixed traffic.** Manual trucks, pedestrians, pallet jacks and the autonomous vehicle share the same floor. The safety case has to cover every one of them, and in Europe it is judged against EN ISO 3691-4, the driverless industrial truck standard. Passing that is a paperwork exercise as much as an engineering one.

- **Variable loads.** A pallet is not a fixed geometry. Overhang, shrink wrap, a broken corner board, a cage with bent feet — the machine has to find the fork pockets anyway, and then verify that what it just placed is actually seated.

- **Consequences.** A warehouse AMR that bumps a shelf drops a tote. A forklift that misjudges a 1.5-tonne load three tiers up drops a stack. Insurance, not enthusiasm, sets the pace of adoption here.

The technical answer to all three is the same: perception plus verification. The truck does not just localise itself, it checks the load, checks the stack, and refuses the operation when the check fails — backing off and calling a human rather than forcing the placement. That refusal is the feature. A machine that only succeeds is a machine nobody insures.

## The incumbents are not standing still

This is not a Chinese company walking into an empty market. Toyota Material Handling, KION's Linde and STILL brands, and Jungheinrich all sell automated trucks, and they have three things a challenger does not: the chassis business, a service network in every European industrial park, and decades of relationships with the people who sign the purchase order.

Their weakness is the same as their strength. Their automation tends to be sold as part of a fleet and a service contract, priced accordingly. The Chinese offer is a retrofit-shaped proposition: a machine that fits a brownfield site, deployed in weeks, supported by a local partner.

There is also a straightforward signal in where these companies spend their marketing budget. VisionNav's autumn calendar included the Tokyo logistics show in September and the Logistics & Automation exhibition in Madrid in November. You do not rent a booth in Madrid to sell to Guangdong.

## Honest limitations

Everything above about performance comes from the vendor. The case studies are published by the company, the deployment count is the company's own, and the ±10 mm figure is a stated accuracy under stated conditions. I have not seen an independent throughput audit, and no neutral party publishes failure rates for autonomous forklifts at scale.

Pricing is opaque across the whole category: nobody publishes list prices, so the common trade-press claim that one side is a third cheaper is an anecdote until you have a quote for your own site.

I have also not covered the part that decides most projects: integration. A forklift fleet has to talk to a warehouse management system that may be fifteen years old, and the cost of that integration frequently exceeds the cost of the trucks. That work is done by local integrators, and their quality varies more than the hardware does.

Finally, service. A truck that stops in a cold store at 2 a.m. needs a technician, not a support ticket. The Chinese vendors are building that network, but outside a handful of countries it is thinner than what Toyota has.

## What readers can do now

- **If you operate a warehouse:** measure before you shop. Aisle width, pallet quality, floor flatness and the ratio of manual to autonomous traffic will eliminate most of the market for you before price is discussed. Then ask for a paid proof of concept on your worst aisle, not a demo in a showroom.

- **If you are specifying the project:** put the refusal behaviour in the contract. Ask what the truck does when the stack check fails, who is alerted, and how long it takes a human to arrive. That answer separates a real deployment from a pilot that never scales.

- **If you are watching the sector:** track installations in Japan and Germany rather than units shipped from China. Those two markets have the highest safety bar and the most entrenched incumbents, so a reference customer there is worth more than a hundred domestic ones.

The driverless forklift is not a spectacular machine. It moves a pallet from one place to another, slowly, and stops when something is wrong. That is exactly why it is likely to be one of the first robots to actually change a factory floor, rather than a demo video.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-robot-autonomous-forklift
Free to quote with attribution and a link to the original.
