In late March 2026, we completed a four-week pilot at a mid-size 3PL facility in Fukuoka's Higashi Ward. The operator runs a regional e-commerce fulfillment operation across roughly 2,200 square meters of picking floor. They had looked at fenced AMR deployments twice before and both times backed out of the project after the floor plan for exclusion zones came back from their facilities contractor. The zones would have consumed about 18% of usable pick floor in their highest-density aisles.
This pilot used a single Closer unit operating in what we call zone-free mode: no physical barriers, no floor tape defining robot paths, no exclusion zones requiring workers to stay clear. The robot shared the pick aisles with between four and eight human pickers depending on shift, day one through to day thirty.
This is a write-up of what happened operationally, what surprised us, and what the operator changed about their workflow over that first week. We are sharing it because the pre-deployment questions we hear most often are not technical. They are operational: "How do staff react?" "What breaks first?" "What do I tell my safety officer?"
Day one: the robot is slow and that is not a problem
We started the unit at 60% of nominal aisle speed for the first three days. This was a deliberate choice, not a technical limitation. We wanted pickers to encounter the robot at a speed that felt obviously stoppable before they experienced it moving faster. At 0.72 m/s the robot moves at a brisk walking pace, which felt about right for staff who had never worked alongside an autonomous unit before.
The reaction of the picking team on day one fell roughly into two camps. About half the staff treated the robot like a piece of furniture that happened to move and found ways to work around it immediately. The other half stopped and watched it pass every time they encountered it in an aisle, at least during the first hour. By the end of the first shift, that "stop and observe" behavior had largely disappeared. People were not watching the robot anymore; they were watching their pick lists.
The one friction point on day one was aisle entry. When a picker approaches an aisle mouth from perpendicular and the robot is already mid-aisle, there is a moment where both need to decide who yields. The robot will detect the picker and decelerate if they step in. But a picker who has not observed the robot's response yet does not know that: they default to standing at the aisle entrance and waiting. In that scenario, the robot decelerates, the picker waits, and both end up stalled until one of them decides to proceed.
We expected this and had briefed the supervisor team to tell pickers: "Walk in normally. The robot will give you space." By day three, aisle entry hesitation had dropped significantly. By day five, we observed pickers walking directly into occupied aisles without breaking stride.
The cage-removal change that actually mattered most
When facilities run fenced AMR routes, the floor space inside those exclusion zones is operationally dead. Shelving cannot be placed there. Pickers cannot cut through them. Pick density in adjacent aisles increases because the productive floor is smaller.
In this facility, removing the planned exclusion zones (the reason two previous deployments had been cancelled) meant roughly 380 square meters of picking floor that would have been sterilized under a fenced deployment remained accessible. That is not a software benefit. It is a floor planning benefit that precedes any efficiency discussion about the robot itself.
The operations manager noted this specifically during week two. The value the robot added was not primarily "it carries things faster." It was "I didn't have to give up floor I was already using." That framing matters for operators evaluating whether cage-free is worth the extra sensor investment over a simpler fenced system.
What changed in pick workflow by day five
By the end of the first week, three operational changes had become informal standard practice among pickers, without any instruction from the supervisor team:
Passing protocol in narrow aisles. In the two narrowest pick aisles (1.8m wide), pickers learned to angle their cart slightly toward the shelving when the robot was oncoming, allowing the robot to pass on the opposite side with about 400mm clearance. They were doing this without being told. It mirrors the behavior people use when passing someone with a trolley in a supermarket.
Reducing mid-aisle stationary time. Pickers who previously paused mid-aisle to check their list started pulling partially to one side before stopping. This was not a robot accommodation specifically: it also made it easier for other pickers to pass. The robot's presence may have catalyzed a general courtesy shift in aisle behavior.
Not rushing to claim priority. In the first day, some pickers would quicken their pace when they saw the robot entering an aisle from the opposite end, trying to complete their picks before the robot arrived. By day three, this had stopped. People trusted that the robot would slow if they were still in the aisle. The trust was built through repeated uneventful encounters, not through training.
The incident log for the first week
No physical contacts with staff. One basket left on the floor in an aisle caused the robot to perform a full stop; it waited for 30 seconds, the basket was not cleared, and the robot executed a re-route to the parallel aisle. One picker brushed the robot's side chassis with their sleeve while passing in the narrow aisle; the robot detected the contact, decelerated to zero, and resumed after the person had moved clear. No injuries, no damage, no panic.
The supervisor flagged one near-miss on day two that was not captured in robot telemetry: a picker rounded an aisle end-cap at speed without looking and came within about one meter of the robot coming from the other side. The robot decelerated to a crawl; the picker startled and stepped back. Telemetry showed the detection occurred at approximately 1.1m distance, deceleration command issued, robot at 0.15 m/s when they were at closest point. No contact. The supervisor used this as the basis for a brief end-of-shift talk about looking before taking aisle corners at speed, which is good practice regardless of robot deployment.
What the safety officer asked about
The facility's safety officer reviewed the first week on day eight. His questions were almost entirely about documentation rather than behavior: Where is the risk assessment? What standard does this comply with? What happens if the robot runs into something and there is a liability question?
We had prepared a written risk assessment for the deployment covering the ISO 3691-4 requirements that apply to autonomous mobile robots in industrial settings, the specific test scenarios we had validated before shipping the unit, and the pilot configuration parameters. The safety officer reviewed it and added two requirements: an end-of-shift robot position log (showing where the robot was when each shift ended, to confirm it was parked in the designated charge bay), and a weekly telemetry summary emailed to him directly.
Both of those were reasonable requests and straightforward to implement. The documentation process took about two hours of our time. What it confirmed is that the safety officer's job is risk assessment documentation, not risk management. The robot's behavior had already passed his informal visual observation. The documentation was the formal sign-off layer.
Week one is not week twelve
Zone-free deployment does not mean zero operational adjustment. The first week requires active observation: watching how staff and the robot interact, noting any friction points that human supervisors can address through informal guidance, and being on-site to answer questions as they come up. We had an engineer on-site for the first three days, then remote access for the remainder of week one.
The more important observation is that the hardest part of removing exclusion zones is not the robot. It is helping the operator and their staff develop a mental model of a machine that shares the floor rather than owns a zone of it. Once that model is in place, the practical adjustments are small and mostly happen organically. The pickers in this facility adapted faster than we expected, partly because the robot's behavior was legible: it slowed when they were close, it gave way, it did not claim priority. It behaved, in the words of one picker we spoke with at the end of week one, "like a polite machine."
That description is not marketing language. It is what zone-free deployment looks like when the sensing and motion adjustment are working as intended.
