Compendium

Walkthrough — Design a Fully Autonomous Fulfillment Warehouse

15 min read

Walkthrough — Design a Fully Autonomous Fulfillment Warehouse

A greenfield 200,000 sqft (18,600 m²) e-commerce fulfillment center processing 50,000 outbound orders per day. Fully integrated robotic fleet: autonomous mobile robots (AMRs), goods-to-person automated storage and retrieval systems (ASRS), piece-picking robotic arms, automated sortation, conveyor belts, put-walls, and the warehouse-management + warehouse-control + warehouse-execution + fleet-orchestration software stack that makes it all flow.

Reference deployments: Ocado CFC, Amazon Robotics FC, Walmart × Symbotic Pacific Coast facility, Target × AutoStore, Kroger × Ocado, Albertsons × Takeoff, JD.com 4S, Alibaba Cainiao, DHL × Boston Dynamics Stretch, Best Buy, American Eagle Quiet Logistics, ARC × Maersk. Per-unit AMR design is covered in design-autonomous-mobile-robot; this walkthrough designs the facility that orchestrates hundreds of them.

1. What we’re building

A two-level steel-framed distribution facility:

  • Footprint: 200,000 sqft (18,600 m²), single building, 35 ft (10.7 m) clear height for ASRS + mezzanine
  • Throughput: 50,000 orders/day (~3000 units/hr peak, ~1500/hr average across 16 operating hours)
  • SKU count: 100,000 active SKUs, ABC distribution (top 20% drives 80% of volume)
  • Order profile: 60% single-line orders, 30% two-to-five line, 10% multi-line. Average 1.8 units/order, average unit weight 0.8 kg (1.8 lb), dimensions ≤30×30×30 cm
  • Service area: regional next-day delivery, 250-mile (400 km) radius
  • Operating model: 16 hr/day across 2 shifts, 7 days/week, ~5800 operating hours/year
  • Workforce: 80 humans (down from 400 in a conventional facility) — predominantly induction, exception handling, supervision, maintenance

The fleet:

  • ~150 AMRs (mixed Locus + MiR + OTTO classes) for cross-dock + put-wall replenishment + person-following
  • 60,000-bin AutoStore-class ASRS goods-to-person grid covering 80,000 sqft
  • 120 piece-picking stations with collaborative arms (Berkshire Grey + Dexterity + Covariant Brain)
  • 4 automated sortation loops (cross-belt + tilt-tray + bombay)
  • 6 manual fast-pick zones with put-walls for low-volume long-tail SKUs
  • 12 dock doors with conveyor-integrated palletizers + label-print-and-apply
  • Stretch (Boston Dynamics) for truck-unload

The integration target: 99.5% on-time order completion, ~12 second cycle from order release to pack station, <0.05% mispick rate, OSHA-recordable injury rate below 1.5 per 100 FTE-years.

2. Spec table

ParameterValueNotes
Floor area200,000 sqft (18,600 m²)greenfield, 50 ft (15 m) bay spacing
Clear height35 ft (10.7 m)enables 2-level mezzanine + ASRS
Orders/day50,000peak factor 1.5×
Lines/hour peak~5400drives sortation sizing
Pick stations120 piece + 20 case + 12 roboticmixed manual + automated
Storage capacity60,000 ASRS bins + 30,000 pallet positionsmixed reserve + active
AMR fleet~150 robots across 4 SKUsLocus, MiR, OTTO, Stretch
Sortation loops4 (cross-belt + tilt-tray + bombay)shipping + zone routing
WMS / WES / WCSManhattan Active + Locus Bridge + custom WCSthree-layer software stack
NetworkWi-Fi 6 (802.11ax) + private LTE/CBRS backupsafety-critical link redundancy
Power8 MVA primary + 2 MVA standby13.8 kV → 480 V → 208 V
HVAC35–75 °F (1.7–24 °C) cooling, ≥40% RHelectronics + cardboard storage
Fire protectionESFR sprinklers + in-rack + smoke purgeNFPA 13 + NFPA 230 storage
CAPEX (greenfield)$80–200 M facility + $30–80 M fleettotal $110–280 M
Payback3–5 years vs manual baselinedepends on labor cost + volume
Operating cost$0.40–0.80 per ordervs $1.20–2.50 manual

The per-line economics is the central business case: at 50,000 orders/day × $0.80 savings per order × 350 days/year = $14 M/yr operational savings, payback on $50 M of automation at ~3.5 years. Higher labor-cost markets (US, Western Europe) pay back faster; lower-labor markets (India, Vietnam, Mexico) typically settle for partial automation.

See mobile-bases + robot-platform-catalog for AMR + ASRS platform catalog and industrial-automation for the WMS/WCS protocol stack.

3. AMR fleet

The AMR fleet is heterogeneous because no single platform optimizes for every task. The fleet mix:

  • Locus Origin / Vector / Max (~80 units): collaborative cart-following AMRs for piece-pick from manual fast-pick zones to pack stations. Per-unit BOM ~$30–40k, lease ~$1500–2000/mo. Specifically Vector for piece + Max for case (heavier).
  • MiR 250 / 600 / 1350 (~30 units, Teradyne): general-purpose payload transport, dunnage, tote relay. MiR 250 carries 250 kg, MiR 1350 carries 1350 kg. $50–95k each.
  • OTTO 100 / 600 / 1500 (Rockwell Automation, ~20 units): heavier payload tug, pallet movement. OTTO 1500 moves a 1500 kg pallet at 2 m/s.
  • Geek+ M / S / RoboShuttle / R-Series + Hai Robotics A42T HaiPick (~15 units): rack-shuttle AMRs that bring entire shelves to pick stations. Geek+ M lifts 1000 kg.
  • Boston Dynamics Stretch (~2 units): articulated-arm truck-unload at ~800 cases/hr per arm. $300–500k each. Stretch is the breakthrough in unstructured truck-unload that previously required exclusively humans.
  • Amazon Robotics Pegasus / Hercules / Proteus / Sparrow (referenced as benchmark only; Amazon doesn’t sell externally): rack-shuttles + first fully-autonomous Amazon AMR (Proteus, 2022).
  • Quicktron Q-Series, Fetch Freight 500/1500 (Zebra), Pixel (alternates for AMR + arm-equipped roles).

Each platform reports to the fleet orchestrator via VDA 5050:2022 (JSON over MQTT). The orchestrator handles task allocation, conflict resolution, charging schedules, and fleet rebalancing. See design-autonomous-mobile-robot for the per-AMR design (drive, sensors, safety, navigation, BOM).

Fleet orchestration software:

  • Open-RMF (Open Robotics + AWS RoboMaker, partly maintained post-AWS-discontinuation in 2024): open-source multi-fleet orchestrator. Used as the integration layer.
  • Locus Bridge: vendor-agnostic fleet orchestrator that abstracts heterogeneous robot APIs.
  • MiR WISE: MiR-native fleet manager; integrates into Open-RMF.
  • NVIDIA Isaac Manipulator + Isaac Perceptor (2024+): simulation + perception infrastructure for sim-to-real transfer of new behaviors.

The orchestrator runs on a redundant pair of Dell PowerEdge R760 servers in the warehouse data closet (RHEL 9, K3s lightweight Kubernetes). MQTT broker: HiveMQ Enterprise with TLS 1.3 mutual cert auth.

4. ASRS — goods-to-person storage

The ASRS is the throughput core. Choice between several architectures, each with different cube-utilization vs throughput trade-off:

  • AutoStore B1: 3D cube-storage grid, ~700 mm × 400 mm × 220 mm bins, robots run on top of the grid retrieving from any column. ~30 bins/hour per port, scales to thousands of ports. ~$25–35M for 60k bins. Best cube utilization (4×). The reference choice here.
  • Symbotic SymBots (Walmart’s choice, ~$50M+): autonomous shuttles in a multi-level rack moving cases not bins. Bigger payload per move, lower bin-density. Walmart’s Pacific Coast facility runs at ~1000 cases/hr per zone.
  • GreyOrange GreyMatter / Butler XL: rack-shuttle goods-to-person.
  • Exotec Skypod: 3D rack with autonomous robots that climb the rack vertically. Highest throughput per port (~700 lines/hr per station). $5–7M per zone of ~10k bins.
  • Geek+ T-series: similar 3D rack with climbing robots.
  • Knapp OSR Shuttle + YLOG-Shuttle: shuttle-per-level systems. Lower per-shuttle cost but more shuttles.
  • Vanderlande FastPick + Pick-it-Easy: classic mini-load AS/RS.
  • Berkshire Grey Robotic Shuttle Put: integrates ASRS shuttles with put-wall sortation.

We pick AutoStore B1 for the storage core — 80,000 sqft of grid, 60,000 bin positions, 200 R5 Plus robots running on top, 12 carousel ports presenting bins to operators or robotic pickers. CAPEX ~$30M for the AutoStore portion alone. Cube utilization advantage means the same SKU count in conventional racking would need ~250,000 sqft — saving 50% of building footprint.

ASRS protocol stack: AutoStore proprietary controller communicates with the WCS over MQTT + REST. Cycle: order release → WCS computes bin sequence → AutoStore controller assigns robots → bins arrive at port within 20–60 sec.

5. Piece pick — robotic arms

The piece-picking automation is the highest-leverage automation in 2025–2026 because piece-picking was the last bastion of human work in fulfillment. Three technology stacks:

  • Berkshire Grey RPSi: Robotic Pick Station with integrated AutoStore. End-effector: suction + finger gripper hybrid. ~700 picks/hr per cell.
  • Covariant Brain (now part of Amazon Robotics 2024): AI-powered piece-pick with continuous learning. Used by Knapp, Ocado, KMW, several others. ~600–900 picks/hr depending on item mix.
  • Dexterity Sortation + Pickle Robot: parcel-induct and depalletizing. Pickle specializes in unstructured-truck unload.
  • Plus One Robotics: parcel induction at scale. Used by FedEx, USPS.
  • Mujin: high-end mixed-SKU induction. Used in Amazon, JD.com.
  • Sapien Robotics + Kindred Sort (Ocado-owned 2020): piece-pick + sort.
  • Soft Robotics: suction + soft-gripper end-effectors for unstructured produce + bakery.
  • Locus PickWise: in-cart pick assist (not full robotic pick, but cobot-style guidance).
  • AutoStore RoboPick: AutoStore’s own pick robot at AutoStore ports.

Our design: 80 Berkshire Grey RPSi cells at the AutoStore ports + 20 Covariant Brain pick stations for higher-mix exception items + 20 Dexterity arms for outbound case-pack consolidation. ~$150–300k per cell installed. Total piece-pick capex ~$30M.

Vision: each cell carries an Intel RealSense D455 stereo depth camera + a Photoneo PhoXi 3D structured-light scanner for high-precision picks. Compute: NVIDIA Jetson AGX Orin per cell (~$2k) for local inference, with central cluster-level training on a NVIDIA DGX H100 in the data closet.

6. Sortation

Outbound sortation routes picked items to the right shipping lane (parcel carrier, store-replenishment route, B2B route, or expedited):

  • Honeywell Intelligrated MaxSort cross-belt sorter: high-throughput (~15,000 packages/hr), gentle for fragile + small. ~$3–5M per loop installed.
  • Vanderlande POSISORTER BS5 / BG Sort: cross-belt sorter, similar throughput.
  • Dematic Multi-Shuttle: tilt-tray sorter for medium items.
  • Beumer LS-4000: cross-belt + tilt-tray combinations.
  • Crisplant LS-4000E: tilt-tray.
  • Equinox Sort: emerging bombay sorter (drop-bottom). Best for irregular items.
  • Witron OPM: tray-based sorter.

We deploy two cross-belt loops (Honeywell + Vanderlande) for parcel + carton, one tilt-tray for irregular, one bombay for high-velocity fast-pick consolidation. Total sortation CAPEX ~$15M.

Sortation is the bottleneck-limiting subsystem — if the sorters can’t keep up, picks back up at the ASRS ports. Sortation is sized at 1.5× peak demand to absorb burstiness.

7. Conveyors + put-walls + induction

Conveyance threads the building together:

  • Powered roller + belt conveyor: ~5 km (3.1 mi) of conveyor across the facility. Honeywell Intelligrated, Hytrol, TGW.
  • Put-walls: ~40 put-wall positions (lighted slot displays per order, operator places items by indicator light). Lightning Pick (Matthews), Bastian Solutions.
  • Pack stations: 30 manual + 12 auto-pack (box-on-demand void-fill + sealing). Sparck Technologies CVP-500 for auto-pack at ~10 sec/order.
  • Print-and-apply: shipping label + return label automation at pack-station exit. Markem-Imaje, Avery Dennison, Domino.
  • Carton sealing: 3M-Matic case sealers at pack-station output.

Total conveyor + put-wall + pack capex: ~$12M.

8. Software stack — WMS, WCS, WES, OMS

Four-layer software hierarchy (with significant vendor overlap between layers):

  • OMS (Order Management System): order intake from web/mobile/marketplace; allocates to facility. Manhattan Active Order Management, Salesforce OMS, IBM Sterling, NetSuite OMS, Veeqo (Amazon-owned).
  • WMS (Warehouse Management System): receives the order from OMS, plans the work, tracks inventory + labor. Manhattan Active Warehouse Management (the de-facto enterprise leader), SAP EWM, Blue Yonder Luminate (formerly RedPrairie + JDA), Körber K.Motion (formerly HighJump), Microlistics, Made4Net, Softeon.
  • WCS (Warehouse Control System): low-level orchestration of conveyors + sorters + ASRS. Often vendor-specific: Honeywell Intelligrated WCS, Dematic iQ, Daifuku, vendor-proprietary on each piece of MHE (material handling equipment).
  • WES (Warehouse Execution System): bridges WMS + WCS, real-time work-flow optimization. Manhattan WES, Softeon WES, Locus, 6 River Systems MFP (Shopify-owned 2019, spun out 2023 to Ocado).

Our deployment stack:

  • WMS: Manhattan Active Warehouse Management (cloud-native, ~$2–4M/yr SaaS at this volume)
  • OMS: Manhattan Active Order Management (same vendor stack saves integration headache)
  • WES: Manhattan WES extension
  • WCS: vendor-bundled with each MHE supplier; integration layer via Manhattan APIs
  • Fleet orchestration: Open-RMF as central + Locus Bridge as vendor-agnostic abstraction
  • Robotics dev infrastructure: NVIDIA Isaac Sim + Isaac Lab for sim-to-real, NVIDIA GR00T humanoid models for emerging Stretch + Pixel + similar (see design-humanoid-full-stack)

Integration: Manhattan APIs (REST + GraphQL) feed work to the WES, WES emits VDA 5050 mission orders to fleet manager (MQTT), fleet manager dispatches to AMRs + arms. End-to-end order-to-tendered cycle: 8–15 minutes for a typical 1.8-line order.

Cybersecurity: full network segmentation, identity-aware access (Okta/Azure AD), 24/7 SOC monitoring (see design-soc-incident-response-platform for the SOC architecture pattern), per-device certificates on all MHE + AMRs.

9. Throughput economics

Lines-per-hour-per-associate is the canonical throughput metric:

  • Manual conventional warehouse: 60–80 lines/hr/associate (walking cart-pick)
  • Voice-pick + RF guns: 100–140 lines/hr
  • Goods-to-person (ASRS): 250–400 lines/hr
  • Goods-to-person + robotic pick: 600–900 lines/hr (no human at the pick station except exception handling)
  • AutoStore CarryPick + AutoStore RoboPick combined: 700–1000 lines/hr per cell

The 5400 lines/hr peak in §1 distributes as:

  • 3500 lines/hr through AutoStore + robotic pick cells (80 cells × ~45 lines/hr each effective)
  • 1500 lines/hr through manual fast-pick zones (12 stations × ~125 lines/hr)
  • 400 lines/hr through case-pick + exception handling

Ergonomic limits per NIOSH guidelines: a single human at a put-wall handles ~150 lines/hr sustainable (limit driven by reach + lift cycles at ~1 every 24 sec). NRF Lean reports + slotting algorithms (ABC + velocity-class + cube-velocity) reduce travel + reach by 20–30% in a conventional layout; in an ASRS layout, slotting reduces robot fleet utilization but doesn’t affect human ergonomics directly.

The slotting optimization runs nightly on the WMS — re-ranks SKUs by recent velocity, repositions bins (low-velocity to deep storage, high-velocity to top of carousel rotation) via overnight rebalance moves.

10. Truck dock + outbound

12 dock doors with:

  • Truck-leveler + dock-seal (Rite-Hite, McGuire)
  • Powered conveyor extends into trailer (Caljan, Joloda) for case-by-case loading
  • Boston Dynamics Stretch for inbound truck-unload: lifts ~800 cases/hr per Stretch, replaces 2–4 humans per dock. $300–500k per Stretch.
  • Auto-palletizer (Yaskawa, KUKA, ABB IRB 660, FANUC M-410iC) for outbound pallet build-up. ~600 cases/hr per palletizer.
  • Stretch wrapper + load-securing automation
  • Pallet conveyor to outbound staging

Per-dock CAPEX (including conveyor + Stretch + palletizer): ~$1.0–1.5M × 12 docks = $15M.

Dock-to-shipping turnaround: 30 minutes per truck (vs 90+ minutes manual). Drive cycles: ~80 trucks/day across 12 docks at peak.

11. Building + utilities

Greenfield building cost in 2025–2026 US tier-2 metro: $300–500/sqft for distribution-grade fit-out (vs $200/sqft for cold dark warehouse).

  • Site: 25–35 acres including parking, drives, retention. Land $15–40/sqft = $15–60M.
  • Building shell: tilt-up concrete + steel structure, 35 ft clear, insulated metal panel walls. ~$80/sqft = $16M.
  • Fit-out (HVAC + electrical + fire + life-safety): ~$120–150/sqft = $24–30M.
  • Mezzanine + conveyance steel: ~$50/sqft mezzanine zone = $5M.
  • Fire protection: ESFR sprinklers (NFPA 13) + in-rack sprinklers for ASRS (NFPA 230) + smoke purge. ~$3–5M.
  • HVAC: 35–75 °F (1.7–24 °C), 40% RH, ~20,000 ton-hr/year cooling. Trane, Daikin, Carrier. ~$8–12M.
  • Power: 8 MVA primary service at 13.8 kV, step-down to 480 V three-phase + 208/120 V branches. Pad-mount transformers + 2 MVA standby generator. ~$5–8M.
  • Networking: Cisco Catalyst + Meraki backbone, ~600 Wi-Fi 6 APs + private CBRS LTE backup, redundant 10 GbE fiber to all MHE controllers. ~$2–3M.

Total building + utilities CAPEX: $80–135M for a 200k sqft facility. Add $15–40M for land.

Sustainability: rooftop solar PV (200,000 sqft × 20 W/sqft = 4 MW) offsets ~30% of facility load — see design-utility-scale-solar-pv-plant for PV design and design-residential-solar-battery-system for rooftop-scale BESS integration.

12. Schedule + program economics

Greenfield program from site selection to first ship:

PhaseDurationDeliverables
Site selection + design6 monthssite + building design, MHE RFP, vendor selection
Permitting + financing4 months (parallel)building permits, environmental review, financing close
Site work + building shell12 monthspad, foundation, tilt-up, roof, dry-in
MHE installation8 months (overlapping shell)ASRS install + commissioning, conveyor, sorters, robots
Software integration + UAT4 monthsWMS go-live, WES integration, sim-trained pickers
Pilot operations + ramp6 months10% → 50% → 100% volume ramp, throughput validation
Total to full operations24–30 monthsfirst commercial revenue at month ~20

Total CAPEX rollup:

ItemCost
Land + building shell + fit-out + utilities$100M
AutoStore + 60k bins + 200 robots + ports$30M
Piece-pick automation (120 cells)$30M
Sortation (4 loops)$15M
Conveyance + put-walls + pack$12M
AMR fleet (150 units)$8M
Dock automation + Stretch$15M
Software (WMS + WES + integration)$5M
Total$215M

Operational economics at 50,000 orders/day × 350 days/year = 17.5M orders/year:

  • Manual baseline: 17.5M × $2.00 = $35M/yr labor + ops
  • Automated: 17.5M × $0.60 = $10.5M/yr labor + ops + maintenance + utilities
  • Annual savings: $24.5M
  • Payback on $215M CAPEX: ~8.8 years gross; ~6 years after labor inflation factored

Higher labor-cost markets (West Coast US, Northeast US, Western Europe) hit 3–5 year payback. Lower labor-cost markets pay back slowly enough that partial automation (manual zones for long-tail SKUs, AMRs for cross-dock + put-wall replenishment only) is the right answer.

13. Reliability + maintenance

System-of-systems reliability is the dominant operational concern:

  • MTBF by subsystem: ASRS robots ~6000 hr, AMRs ~5000 hr, conveyor motors ~30,000 hr, robotic pick cells ~2000 hr.
  • Spares strategy: hot-swap robots + cell components; on-site spare pool of 10% of AMR fleet + 5% of ASRS robots.
  • Predictive maintenance: vibration + thermal sensors on conveyor + sorter drives, ingestion to a time-series DB (InfluxDB, Timescale), ML anomaly detection.
  • Maintenance staff: 12–18 maintenance technicians (24/7 coverage in 2-shift mode), 2 controls engineers, 1 site reliability lead.
  • Vendor service contracts: ~10% of MHE CAPEX/yr ongoing service ~$10M/yr.

OEE (overall equipment effectiveness) target: 85%+. Below 75% triggers a structured root-cause analysis (5 Whys + Ishikawa). Industry-leading FCs hit 90%+ OEE on the goods-to-person subsystem.

14. Reference deployments

For benchmarking the design:

  • Ocado CFC (Customer Fulfillment Center): ~10 facilities globally including Erith UK, Andover UK, Atlanta US (Kroger), Charlton UK. Hive (cube ASRS) is the equivalent of AutoStore — Ocado developed in parallel.
  • Amazon Robotics FCs: 350+ globally as of 2025. Pegasus (drive-and-sort), Hercules (heavy lift), Proteus (first fully autonomous Amazon AMR, 2022), Sparrow (piece-pick arm, 2022).
  • Walmart × Symbotic: Pacific Coast facility + 42+ regional DCs being retrofitted. SymBot autonomous case-shuttles.
  • Target × AutoStore: regional sortation centers, ~10 facilities.
  • Kroger × Ocado: 20+ “sheds” rolling out 2023–2027.
  • Albertsons × Takeoff + Albertsons × Symbotic: hybrid micro-fulfillment + central.
  • JD.com 4S: Shanghai “Asia No. 1” + 28 more across China, mixed Geek+ + JD-built robots.
  • Alibaba Cainiao: Hangzhou, Wuxi, more — Quicktron + Hai Robotics + Cainiao-built.
  • DHL Smart Robotics + Boston Dynamics Stretch: Stretch deployment across DHL Supply Chain 2023+.
  • Best Buy + American Eagle (Quiet Logistics) + ARC Maersk: niche but representative mid-tier rollouts.

15. Adjacent

End walkthrough — design-autonomous-warehouse.md.

Graph View

Backlinks