How a senior production manager overseeing multi-vendor AGV/AMR fleets across large-scale manufacturing buildings discovered the structural limits of today's fleet management systems — and why the industry needs a fundamentally different approach to robot traffic control.
The Scale Few People Talk About
When most people think of warehouse robots, they picture a tidy Amazon fulfillment center with uniform orange pods gliding in neat rows. The reality inside Japanese manufacturing facilities tells a very different story.
We recently spoke with a senior production manager at a major Japanese manufacturer — someone who oversees the construction and automation of manufacturing buildings where potentially 100 autonomous robots operate per building. His company is in the middle of a multi-year, multi-billion-yen investment cycle to significantly increase labor productivity, building entirely new factories while expanding existing ones.
What he described was an automation landscape far more complex, messy, and instructive than any demo video suggests.
The Architecture: MES to PLC to FMS to Robots
The first thing that stands out is how deeply robot control is embedded within the manufacturing execution architecture. Unlike a standalone logistics warehouse where a fleet management system (FMS) might sit at the top of the stack, this facility operates with a layered control hierarchy.
At the top sits a Manufacturing Execution System (MES) that issues manufacturing instructions and enforces full traceability using RFID tags on every material. This level of traceability is standard in highly regulated manufacturing environments.
Below the MES sits a PLC layer using controllers from two of Japan's dominant PLC manufacturers, connected via CC-Link — an industrial communication protocol widely adopted in Japanese manufacturing. The PLCs handle interlocking and handshake signals between different pieces of equipment. They maintain shared memory flags that control zone-based entry prohibition for robots.
Below that, each robot vendor's FMS manages its own fleet. The robots themselves include:
- Magnetic-guided AGVs
- SLAM-based autonomous mobile robots (AMRs)
- Co-developed custom robots built through joint development programs with partner companies
- Multi-axis industrial robots handling specialized process tasks
The result: a heterogeneous fleet where 3-4 different robot types operate in one production area (~10,000 m²), and 5-10 different types across the larger processing and packaging zones (~25,000+ m²).
The Multi-Vendor Reality
This was one of the most candid admissions in our conversation. When pressed on what "struggling" means in practice, the manager described a system where:
- Different vendors' FMS cannot share real-time position data with each other. Each fleet operates in its own silo.
- PLC shared memory serves as the coordination layer. Entry prohibition flags are maintained in PLC shared memory, and CC-Link signals handle the traffic control between zones.
- System integrators mediate the connections. Major Japanese electronics and industrial companies handle the "relay" work between vendors — essentially acting as translators between incompatible fleet management systems.
The company's standard approach is to order from a single major logistics provider — warehouse and transport equipment bundled together — precisely to minimize multi-vendor integration headaches. But manufacturing reality does not cooperate. Legacy equipment from co-development programs in the 2000s and 2010s coexists with newer commercial systems. Different processes have different physical requirements.
The result is vendor diversity whether you want it or not.
Congestion: Daily, Inevitable, and Sometimes Catastrophic
We asked directly: how often do congestion and deadlocks occur?
And then, the statement that stopped us: we asked whether it is possible to build a factory where congestion never happens.
"No. That doesn't exist."
He paused, reflected, and repeated: "No, there really isn't such a thing."
Root Causes (Ranked from Interview)
| Rank | Root Cause | Details |
|---|---|---|
| 1 | PLC program bugs | The most common cause. "What stops things is basically PLC program bugs." The inner workings of vendor-provided PLC programs are a black box to the user side. |
| 2 | SLAM robot standoffs | Multiple SLAM-based robots meeting in a corridor and freezing, each waiting for the other to move. A classic multi-agent deadlock. |
| 3 | Human entry into robot field of view | Triggering emergency stops in areas where people and robots coexist. |
| 4 | Narrow corridors | Building structure constraints create bottlenecks that no amount of software tuning can solve. |
| 5 | Wet floors | Washing and cleaning processes leave moisture that disrupts SLAM map matching, causing robots to get stuck. |
| 6 | Co-developed robot defects | Custom robots built through joint development programs are the most problematic, requiring enhanced maintenance contracts with annual service schedules. |
The Worst Case: 3-4 Days of Downtime
Semi-annual major shutdowns are a reality. The most recent one lasted 3-4 days — requiring the logistics vendor to dispatch engineers from a remote location, with travel time adding significant delay before recovery could even begin.
During these episodes, production schedules must be adjusted — a serious matter for a manufacturer whose products are critical to end users.
Regulation Is Not the Barrier You'd Expect
One of our initial hypotheses was that industry-specific quality regulations would create insurmountable barriers to deploying new software in these environments. The answer surprised us.
The real regulatory burden falls on the MES and process control layers — the systems that ensure batch integrity, traceability, and product quality. Robot transport, while important for productivity, sits in a different regulatory category.
Change management procedures exist and must be followed when modifying equipment, but they are not the showstopper we anticipated. The real physical barrier is more prosaic: AGV tires abrading floor coatings, causing material to peel off and create contamination risk. The solution? Replacing entire floor sections with stainless steel — a far more expensive fix than any software deployment.
The Budget: Who Decides, and When
The senior production manager we spoke with is himself the budget planner for mechanical systems, including transport automation. For major investments, executive-level approval is required. System architecture decisions involve a digital solutions department for MES-level and above, but the fundamental design philosophy and vendor selection for mechanical systems — including robots — flow through his team.
Timing is critical to understand:
| System Type | Vendor Selection Lead Time |
|---|---|
| Warehouse systems | ~1 year before construction starts |
| Manufacturing buildings | ~3 years before completion |
His company currently has a new factory in the planning stages, which means vendor selection for that facility's automation systems is actively underway.
What This Tells Us About the Market
1. The Pain Scales Exponentially with Fleet Size
Small fleets (under 15 units) can manage congestion through operational workarounds — aisle widening, floor markings, better training. Medium fleets (15-30 units) can partially mitigate through free-location strategies and zone separation.
But at 100+ units per building with 5-10 different vendor types? The problem becomes structural. No amount of parameter tuning within individual FMS systems can solve cross-vendor coordination when the systems fundamentally cannot share real-time position data.
2. Static Design-Time Controls Have Hit Their Ceiling
The current approach — PLC shared memory flags, zone-based entry prohibition, CC-Link signal handshakes — is essentially a set of traffic lights. It works for preventing the worst collisions, but it cannot dynamically optimize flow.
3. The Ordering Structure Creates Lock-In
In Japanese manufacturing, orders frequently flow through super general contractors (super-zenecons) who bundle construction, equipment, and systems. This means the end user often has limited visibility into individual AGV/AMR pricing and limited ability to introduce third-party software after the initial system is deployed.
Any new solution must either integrate at the initial design stage or demonstrate such compelling value that it justifies reopening a completed procurement package.
4. The Interoperability Standards Are Coming
Our interviewee showed immediate interest when we discussed VDA 5050 — the German interoperability standard for AGV/AMR control interfaces. The concept of a vendor-agnostic communication layer resonated strongly with someone living the multi-vendor coordination challenge daily.
This suggests the Japanese market is ready for standardization, even if adoption lags behind Europe. Solutions that position themselves as standards-compatible will have a significant advantage.
The Question No One Has Answered Yet
When we described our research into centralized multi-agent path finding (MAPF) — algorithms that can coordinate routes for all robots from a single system — the response was immediate and telling.
He then added, unprompted: "This seems like it could be the thread that unravels the problem."
And later, after learning more: "I'm very interested in this. There could be business here going forward."
Key Data Points
| Metric | Value | Source |
|---|---|---|
| Robots per manufacturing building | ~100 units | Direct statement (interview) |
| Robot vendor types (per production area) | 3-10 types | Direct statement (interview) |
| Total floor area (manufacturing building) | 35,000+ m² | Direct statement (interview) |
| Congestion frequency | Several times daily | Direct statement (interview) |
| Recovery time per incident | 30 min - 1 hour | Direct statement (interview) |
| Major shutdown frequency | ~Once per 6 months | Direct statement (interview) |
| Worst-case shutdown duration | 3-4 days | Direct statement (interview) |
| Manual intervention frequency | Several times weekly | Direct statement (interview) |
| Can congestion-free factory be built? | "No" (definitive) | Direct statement (interview) |
| Productivity target | Significant labor productivity increase (multi-year plan) | Direct statement (interview) |
| Investment scale | Multi-billion-yen range | Direct statement (interview) |
| Vendor selection lead time (warehouse) | ~1 year before construction | Direct statement (interview) |
| Vendor selection lead time (manufacturing) | ~3 years before completion | Direct statement (interview) |
| Regulatory barrier for AGV software | Low ("not related to product quality") | Direct statement (interview) |
This interview was conducted as part of Rovnou's ongoing research into robot fleet coordination challenges in Japanese manufacturing and logistics environments. The interviewee's identity and employer have been anonymized at their request. All data points are sourced directly from the recorded interview transcript.
Rovnou is developing multi-agent path finding (MAPF) software that coordinates heterogeneous robot fleets without replacing existing fleet management systems. To learn more, visit rovnou.com.