AGVs and AMRs are the fastest-growing segment of warehouse automation - the most deployable, scalable, and risk-managed entry point. Unlike AS/RS or sortation systems, an AMR fleet can be deployed in phases, scaled seasonally, and redeployed across facilities. For DC applications, AMRs dominate because they adapt to the realities of a live warehouse.
AGV vs. AMR: The Critical Distinction
These terms are used interchangeably in vendor marketing. They are meaningfully different.
| Characteristic | AGV | AMR |
|---|---|---|
| Navigation | Fixed path (wire, magnetic tape, QR code grid) | Free navigation (SLAM) |
| Obstacle response | Stop and wait | Navigate around |
| Path changes | Requires physical infrastructure change | Software update only |
| Installation | High (floor infrastructure) | Low (map + software) |
| Cost | $50,000-150,000/unit | $30,000-100,000/unit |
| Best for | Fixed, repetitive, high-precision transport | Dynamic warehouse environments |
[!practitioner-note] Cost ranges reflect 2024-2025 market estimates based on practitioner sourcing experience. Autonomous forklifts run $60K-$250K+. No industry-wide cost survey confirmed - verify current pricing directly with integrators before using in proposals.
AGVs are appropriate for fixed, controlled environments - automated production lines, fixed dock-to-conveyor routes, manufacturing logistics. AMRs dominate in distribution because they adapt to obstacles, layout changes, and seasonal reconfigurations.
AMR Use Cases
Use Case 1: Tugger AMR (Cart Towing)
AMR tows a cart train of totes, orders, or cartons.
Application: Replenishment from reserve to pick zones; order totes from pick to pack; zone-to-zone transfer in large facilities. Performance: 2-6 carts per train; each cart holds 4-12 totes; 2-4 mph operational speed in a live warehouse (0.5-1 mph is docking approach speed only; Seegrid S7 max = 4.0 mph). ROI driver: Eliminates pedestrian transport - one AMR-tug replaces 1-2 FTE.
Use Case 2: Shelf-Lifting AMR (Goods-to-Person)
AMR drives under mobile shelving, lifts the unit, and delivers to a stationary picking station.
Key vendors: Amazon Robotics (formerly Kiva), Geek+, Hai Robotics, 6 River Systems. Performance: 300-600+ picks/hour at stationary station vs. 80-150 picks/hour walk-and-pick. 2-3× improvement typical. Requirements: Dedicated AMR grid storage area; no mixed use of shelving; Floor FF50+; 4-5 ft aisle widths.
Use Case 3: Pick Assist / Follow-Me AMR
AMR follows an operator, carrying the pick cart. Operator walks and picks; robot carries the load.
Key vendors: 6 River Systems (Chuck), Locus Robotics (LocusBot), Fetch Robotics (Zebra). Performance: 15-30% productivity improvement vs. manual cart. More modest than G2P - operator still travels. Best fit: Operations transitioning from manual; large facilities where cart movement is significant; mixed human/robot environments where full G2P isn’t justified.
2026 deployment data — DHL/Robust.AI Carter: DHL Supply Chain’s five-year Robust.AI Carter collaborative mobile robot program shows 30% productivity improvement in initial Mexico deployments, scaling to 60%+ in mature North American operations — consistent with this range at ramp-up vs. steady-state. Marks the first enterprise cobot deployment in DHL’s Latin American network (Feb 2026); scale target is hundreds of Carter robots across Americas operations. (Source: Research - DHL Robust.AI Carter Cobots Latin America 2026)
2026 update - Locus Array: Locus Robotics launched Locus Array at MODEX 2026 (April), a tower-style mobile robot with an integrated picking arm and AI perception that handles picking, putaway, induction, and replenishment autonomously within standard racking - no infrastructure redesign required. This is a distinct product from the LocusBot pick-assist AMR: Locus Array targets end-to-end autonomous fulfillment (“physical AI” category). Vendor claim: 90% labor reduction. DHL Supply Chain is an early adopter in live production. See Research - Locus Array Physical AI Autonomous Fulfillment 2026.
Use Case 4: Autonomous Forklift (AMR-Forklift)
Fully autonomous pallet handling - picking from floor, rack, or dock and transporting to defined destination.
Key vendors: Seegrid, Vecna Robotics, Balyo (Linde), Toyota (Autopilot), Jungheinrich. Performance: 20-35 pallet moves/hour (vs. human 50-65/hr) - but available 24/7. Economics work: 24/7 operations with expensive overnight labor; hazardous environments (cold storage, narrow aisles); highly predictable, repetitive pallet flows. Don’t work: Dynamic high-congestion environments; poor pallet quality; operations requiring human judgment for exceptions.
Use Case 5: Swarm Sortation AMR (3D Parcel Routing)
Small robots navigate a multi-level 3D rack structure, routing parcels to sort chutes or destinations without conveyor infrastructure. A central swarm algorithm assigns parcels to robots and optimizes paths in real time as units move in parallel.
Key vendor: Unbox Robotics (Pune, India, founded 2019). Source: unbox-robotics-company-profile-2026.
UnboxSort specs (all vendor-reported; no independent audit published as of October 2026):
| Metric | Vendor Claim |
|---|---|
| Throughput | 1,000-20,000 parcels/hr (modular); newsroom: 15,000+ live |
| Sort destinations | 2,000+ |
| Grid height | Up to 2.4 m (multi-level vertical) |
| Sort accuracy | 99.9% |
| Uptime | 98.5% |
| Space savings | 50-70% vs. conveyor systems |
| Productivity vs. manual | 3-5x |
Navigation: Proprietary ACO (Ant Colony Optimization) swarm algorithm. Robots reroute around congestion dynamically. Not SLAM-based free warehouse navigation; robots operate within a defined grid structure.
Pricing: RaaS primary.
Markets: India (primary), US (ProMat 2023 launch), Europe (Parcel+Post Expo 2026), Australia (CeMAT 2026 via Barcomation Group).
Funding: $28M Series B, January 2026 (3one4 Capital). FY2025 revenue growth 5x year-over-year (self-reported).
Positioning: Fills the gap between manual sorting and full conveyor AS/RS. Targets brownfield, mid-market e-commerce, postal/express, and 3PL operators who need scalable sortation without a conveyor tearout.
Primary competition in sortation: Geek+ (global, $281M raised July 2025), GreyOrange (enterprise), Addverb (India), Knapp/Vanderlande (Europe/NA full-sortation AS/RS).
Key limitations:
- Sortation only: no pick-assist, G2P, or pallet transport.
- All performance figures are vendor or investor-sourced; no third-party benchmark or SLA data published as of October 2026.
- Smaller global integration network vs. established players.
- Named customers (Myntra, DTDC, Miroglio Group, Inditex) listed with no independently verified deployment data.
[!practitioner-note] When a client asks about “Unbox Robotics AMR” confirm the use case first. UnboxSort competes with conveyor sorters and crossbelt systems, not with Locus/6 River pick-assist AMRs. The question is almost always about sortation throughput and space savings, not warehouse transport.
Fleet Sizing: The Queueing Model
For detailed fleet sizing math - Little’s Law, cycle time decomposition, charging strategy, and the 70% utilization limit - see AMR Fleet Sizing.
The critical rule not in most vendor proposals: AMR throughput does not scale linearly past 70% utilization. The M/G/1 queueing model shows queue wait at 70% utilization = 2.33× mean service time; at 85% = 5.67×. Path congestion compounds on top. Effective throughput peaks at 65–75% utilization and declines past that.
Design at ≤70% utilization at peak hour. If sizing math produces 85%, you have an undersized fleet - add robots.
The 15% maintenance float is not optional: spec’d fleet = operational minimum ÷ 0.85. Without it, the first two robots entering scheduled maintenance push below throughput commitment.
Fleet Sizing Methodology
Step 1: Define the Task
| Parameter | Input |
|---|---|
| Transport route | Origin → Destination; distance (feet) |
| Frequency | Trips per hour at peak |
| Load per trip | Totes, pallets, or cart trains |
| Cycle components | Load, travel, unload, return |
Step 2: Calculate Cycle Time
Cycle Time = Load Time + Travel Time (loaded) + Unload Time + Return Time
Travel Time = Distance (feet) ÷ Speed (feet/min)Speed assumptions:
- AMR in live warehouse: 150-250 FPM loaded (2.0-3.0 MPH)
- Apply 85-90% reduction for traffic, intersections, acceleration
- Effective speed: 130-220 FPM
[!practitioner-note] The 150–250 FPM live-warehouse operating range is a practitioner estimate. Manufacturer maximum speeds are typically 1.5–2.0 m/s (295–393 FPM); live warehouse throughput modeling conventionally applies a 60–80% utilization factor for dynamic obstacles and safety zones per ANSI/ITSDF B56.5. No independent field study confirming the 150–250 FPM range specifically in production warehouses was located. Use manufacturer specs as anchor and apply the utilization factor for your specific layout.
Example - replenishment run (reserve to pick zone, 300 ft):
| Cycle Element | Time |
|---|---|
| Load cart at reserve | 2.0 min |
| Travel to pick zone (300 ft ÷ 165 FPM) | 1.8 min |
| Unload at pick zone | 1.5 min |
| Return to reserve | 1.5 min |
| Total cycle time | 6.8 min |
Step 3: Required Fleet Size
Robots = (Required Trips/Hour × Cycle Time in Hours) ÷ Target Utilization (80%)
Example:
25 trips/hr × 0.113 hr ÷ 0.80 = 2.83 → round up to 3 robotsAdd 1 robot for maintenance/charging buffer in a 3-robot fleet. Final fleet: 4 robots.
Step 4: Battery and Charging
- Battery life: 8-16 hours depending on load/travel intensity
- Charging stations: 1 per 2-3 robots (opportunity charging model)
- Infrastructure cost: $3,000-8,000 per charging station
- Power draw: 1-3 kW per AMR during charging
Path Design and Traffic Management
Map development (SLAM process):
- Drive robot through facility to build initial map
- Remove dynamic objects (forklifts, temporary storage)
- Validate permanent infrastructure as fixed obstacles
- Program no-go zones (dock areas during active loading, pedestrian zones)
Aisle width minimums:
| AMR Type | Minimum Aisle Width |
|---|---|
| Shelf-lifting (small) | 4.5-5.5 ft |
| Cart-towing tugger | 6-8 ft |
| Pallet AMR-forklift | 8-10 ft |
| Mixed human/AMR | +2-3 ft beyond single-use minimum |
Traffic management: Fleet Management Software (FMS) handles collision avoidance and right-of-way. One-way robot aisles eliminate congestion where possible. Humans always have right-of-way; AMRs stop when humans enter their path.
Payback Benchmarks
| Scenario | Typical Payback |
|---|---|
| AMR picking assist, 3 shifts | 8-12 months |
| AMR picking assist, 1 shift | 20-30 months |
| Shelf-lifting G2P (3 shifts) | 12-18 months |
| Tugger AMR (labor redeployment) | 10-18 months |
[!practitioner-note] Payback periods are field estimates based on typical multi-shift e-commerce deployments with labor redeployment assumptions at $18-22/hr fully-burdened. Actual payback varies significantly with labor rate, shift structure, utilization, and AMR system cost. Prologis (2024) documents a broader 14-24 month range for optimal conditions. Additional references confirming the 8-12 month range for picking-assist AMRs in multi-shift operations: Reemanbot (2026); CXTMS (2026). Note: the payback table distinguishes picking-assist (follow-me) AMRs from shelf-lifting G2P systems — these are different deployment models with different capital costs and productivity multipliers.
Shift utilization is the most sensitive variable. The same fleet achieves payback in 8-12 months at 3 shifts and 20-30 months at 1 shift.
Documented: AMR deployments with 250%+ ROI in multi-year live deployments; 42% 5-year OPEX reduction with labor redeployment.
Cobot market (H1 2026): Research - Cobot Market Recovery China Dominance H1 2026 — recovery from 2024 decline, China’s growing market share, shift to higher-payload cobots.
Integration and Cultural Failure Mode
The most common AMR failure is cultural, not technical. Workers who distrust the system override it, block paths, or move pods the robots are managing - destroying throughput and ROI.
Best practices:
- Pre-launch briefings explaining what robots do and don’t do, and how to interact safely
- Floor markings distinguishing robot paths from human paths
- Clear escalation protocol when a robot is stuck - workers need a way to report, not stand next to a stopped robot
- Feedback loop for operators to flag operational issues