Bottom line first: an RCS fleet expansion from 20 to 80 AMRs can be done without stopping production — but only if you expand the network, the map, and charging before you add a single vehicle. Dump new robots straight into a live fleet and you will be debugging at 3 a.m.

I have watched this go wrong three times, and the script never changes. The most dangerous part of an RCS expansion is not the new robots — it is the old map. The no-go zones and traffic rules that worked fine for 20 vehicles turn into deadlock machines at 80. A narrow aisle that ran fine as an informal one-way lane with 20 bots becomes a six-robot standoff at an intersection that no human supervisor can untangle fast enough. Everyone watches the new vehicles; the veterans watch the map. That is the single most counterintuitive lesson of a fleet cutover: the hardware is the easy part, the map is the bomb.

Two projects taught me this the hard way.

In September 2024, a 3C accessories warehouse in Suzhou hired me to grow their fleet from 24 to 72 AMRs. The only requirement: not a single daytime order could stop. My plan kept the new vehicles off real tasks for the first 48 hours — a "shadow fleet" running empty routes to learn the map — then a staged ramp: 25% in week one, 50% in week two, full fleet only after the health metrics held. The whole RCS cutover took two weeks. Zero downtime. The timeline: new vehicles arrived September 2, the 48-hour shadow fleet ran September 4–5, the first wave of 18 vehicles took real tasks on September 9, the fleet hit 50% on September 16, and a 5 p.m. health report decided each next step. Two weeks later all 72 vehicles were live — and daytime order on-time rate had actually improved from 96% to 98%.

In 2023, a food distribution center in California did the opposite. The client plugged 40 new vehicles into RCS in one shot and never expanded charging. That evening the new bots queued at the chargers and crowded out the charging windows of the old fleet. At 6 a.m. the next morning the entire fleet was flat. The owner's exact words on the phone: "I added robots, not problems." The post-mortem found a charger-to-vehicle ratio of 1:18 — far below the safety line. I had flagged charging before the new vehicles arrived, and the client said "let's get running first." What got running was the charging queue.

A fleet expansion that survives follows four phases, each with a checkable deliverable.

Phase 1: Network and Map First (2–4 Weeks Before the Cutover)

  1. Stress the network. RCS heartbeats and task dispatch for 80 vehicles run more than four times the traffic of 20, and Wi-Fi roaming handoff points must be retested. My approach: before the new vehicles arrive, push the existing 20 vehicles plus simulated traffic until the access points are saturated, and watch for packet loss. If RCS heartbeat loss exceeds 1%, add access points before talking about new vehicles.
  2. Redraw the map. One-way lanes, no-go zones, and charging areas must all be redrawn for an 80-vehicle scale: narrow aisles become one-way, intersections get yield points, charging moves from the corners to the main aisles. Reusing the old map is a failure notice — traffic rules built for 20 vehicles cannot carry the traffic density of 80.
  3. Retune RCS parameters. Heartbeat timeouts and task-retry intervals were configured for the 20-vehicle era and must be re-tuned for 80. My practice: loosen the heartbeat timeout by 20% first, then tighten it once the ramp stabilizes. Every parameter change goes on a change log — who changed it, what changed, and what the rollback value is — and the night-before checklist verifies each one.

Phase 2: Charging Infrastructure (1–2 Weeks Before)

The charger-to-vehicle ratio I have seen in real projects usually lands between 1:8 and 1:10; warehouses with dense task loads need to push toward 1:6. It depends on battery runtime and task intensity — the California site died on exactly this point. Power up the new chargers a week early and confirm the charging curves are healthy. During the expansion, stagger charging windows: offset the new vehicles' charging by 30 minutes from the old fleet so the charging area never turns into a queue that blocks the aisles.

Phase 3: Staged Ramp-Up (1–2 Weeks)

Bring new vehicles into the fleet in three waves. Wave one is 25% of the new vehicles, running 48 hours empty to learn the map before touching real tasks. Wave two takes the fleet to 50% while you watch three metrics: average task completion time, deadlock count, and charging queue time. Full fleet only if none of them degrade. During this phase, publish a daily fleet health report, and the next wave starts only after the site owner signs off. A staged ramp typically runs 1–2 weeks; large sites with complex tasks can stretch to three, and that is fine. Write the rollback triggers down and pin them on the control room wall: more than 3 deadlocks in a day, average task completion time 15% slower than baseline, or charging queues over 20 minutes — any one of them pauses the next wave and rolls the current one back. Thresholds live on the wall, not in someone's head.

Night-Before Cutover Checklist

  • Network: access points pass full-load stress test, RCS heartbeat loss under 1%
  • Map: one-way lanes and no-go zones redrawn for 80-vehicle scale, new vehicles have run every route empty
  • Charging: new chargers powered and tested for 7 days, staggered charging configured in RCS
  • Traffic rules: deadlock manual-takeover SOP drilled, right-of-way priorities updated
  • Rollback plan: each wave has a defined rollback point; new vehicles can be pulled from the fleet and dropped back to manual handling
  • Ramp metrics: baselines for task time, deadlocks, and queue time recorded; any metric crossing its threshold triggers immediate rollback

AMR fleet operating in warehouse aisles

Four Classic Ways a Fleet Cutover Fails

  1. Adding everything at once. There is no "one-shot" RCS expansion. I have seen two projects try to add dozens of vehicles in a single night; both blew up before dawn.
  2. Skipping the map redraw. No-go logic from the 20-vehicle era becomes a deadlock factory at 80.
  3. Charging as an afterthought. Vehicles arriving before chargers just builds you a parking lot of robots queuing to charge.
  4. No rollback plan. If the ramp has no way to pull a bad wave back, the only option at 2 a.m. is to sit there and take it.

Remember this: a fleet expansion is not won by having more vehicles — it is won by having a map, a network, and charging that can carry more vehicles. Expand those three first, add the vehicles second, and zero downtime is the natural result. Do it the other way around, and the midnight phone calls will find you.

By Lin Xiao, WMS Implementation Consultant