Autonomous forklifts: what a warehouse pilot needs to prove

Suppose a warehouse wants an autonomous forklift to carry pallets from a staging area to a marked handoff point. In this fictional pilot, a successful demonstration would show one pickup and delivery. Buying more trucks requires evidence that the same job works during the warehouse’s normal operating conditions, with an acceptable burden on the people supporting it.

Choose a job small enough to evaluate

VisionNav describes its VNP 15 pallet stacker as working with varied carriers, including pallets and storage cages, and connecting with a warehouse management system, or WMS. That product description doesn’t establish compatibility with every pallet or a working connection to your own software.

For the fictional pilot, agree on the permitted pallet and load dimensions, weight limits and load stability requirements with the supplier. Have it confirm the actual vehicle configuration against the proposed task. The floor condition and route clearances belong in that discussion, along with the space needed to pick up and place a load.

Keep truck unloading and high-bay storage outside this first trial. They introduce different conditions; succeeding on a floor-to-floor route would not demonstrate either capability.

Define what happens at the destination, too. A truck that reaches a full handoff area may have nowhere to put its load. Decide who clears that space and how the system pauses new work when the receiving area can’t accept it.

Assess the route before live testing

A sensor demonstration can’t settle whether the warehouse layout gives people enough protection.

The UK Health and Safety Executive’s pedestrian separation guidance recommends separate pedestrian and vehicle routes wherever possible. It explicitly addresses automatic, driverless vehicles on shared routes: provide sufficient clearance and avoid fixtures that could trap someone. A marked walkway needs to reflect where people actually walk, including routes through doors and crossings.

US OSHA’s forklift pedestrian guidance also identifies collisions and falling loads as hazards and recommends separating traffic where possible. It mainly addresses conventional forklift operation, so it isn’t an autonomous-vehicle certification procedure.

Before any live movement, have competent site safety staff and the supplier assess the installation under the applicable local requirements. They should set the permitted conditions and protective measures, including controlled crossings where necessary. Agree who can halt the trial and who may authorize a restart after a fault or obstruction.

Use controlled checks and simulation where appropriate, under that team’s direction. Never put someone in the truck’s path to test obstacle detection. A quiet-hour run also leaves the shift-change traffic problem unanswered.

Count completed movements and recovery work

VisionNav says its robot control system manages tasks and abnormal information, alongside vehicle monitoring. Ask the supplier for the installation’s task and exception records, then check whether its WMS records match physical movements.

For this pilot, define completion as the right pallet at the right handoff point, placed as agreed, with the receiving system recording the correct result. A dispatched instruction or a truck arriving nearby doesn’t meet that definition.

Suppose the team offers 100 jobs within the agreed operating limits. In this illustrative example, 95 finish correctly, but five of those need staff rescue. That gives 95% correct completion and 90% unassisted correct completion. Both figures belong in the report, alongside how much staff time the rescues consumed. Neither is an industry target or evidence from an actual trial.

Keep halted, rejected and cancelled jobs visible rather than silently removing inconvenient work from the denominator. Record charging and waiting time, and distinguish a protective stop from a fault. Don’t reward throughput improvements that depend on weakening protection.

When a connection fails after a pallet moves, staff need to reconcile its physical position with the recorded job state before resuming work. Define that recovery procedure with the supplier; don’t assume that resending an instruction will be safe. Name the person responsible for an unresolved movement, including outside normal support hours.

Compare equivalent work before estimating savings

VisionNav’s homepage, checked on 10 October 2026, advertises a 30% efficiency increase and a one-to-three-year return on investment. Treat those as supplier claims. They don’t calculate the outcome of this fictional warehouse’s route.

Measure the existing process on comparable loads and shifts before judging the pilot. Include the time people spend preparing pallets, clearing the destination or recovering a failed job. Count work that moves onto another employee’s shift instead of disappearing. A fast vehicle can still wait at the handoff point.

Ask for installation and integration costs, along with the layout changes and protective measures the site assessment requires. Budget for training and ongoing support, including maintenance. Time released for other work may create useful capacity without reducing payroll; converting it into cash savings requires an actual staffing decision.

The pilot report should state which jobs the truck completed, under what conditions, and with how much human assistance. Use those results to decide whether to keep the route, change the setup or propose a wider trial. HSE’s workplace transport risk assessment guidance calls for reassessment before significant changes to vehicles, routes or work. Expanding the pilot needs that review as well as a new operating target.