Warehouse and logistics robots

Warehouses and distribution centres have become one of the fastest-growing markets for robots. Online retail made order profiles smaller and more numerous, customers came to expect next-day or same-day delivery, and labour for repetitive walking and lifting became scarce and expensive in both Canada and the United States. Logistics robots now move pallets and totes, bring shelves and bins to stationary workers, pick individual items, sort parcels, load and unload trucks, and build mixed pallets. ROB-340 covered the localization, mapping and navigation that mobile robots use, ending with a SLAM system for a warehouse robot. This lesson looks at the logistics side: the types of systems, how they are sized, how fleets are managed and integrated with warehouse software, and the safety standards that apply to industrial mobile robots.

The logistics automation landscape

Warehouse operations follow a basic flow: receiving goods from suppliers, putting them away into storage, picking items for customer orders, packing, sorting and shipping. Picking is usually the most labour-intensive step. In a traditional person-to-goods warehouse, pickers walk or drive through aisles with a list, and walking can account for half or more of their time. Automation attacks this in two ways: by moving goods to stationary people, called goods-to-person, or by automating the picking itself.

AGVs and AMRs

AGVs have been used since the 1950s. Early vehicles followed wires buried in the floor; later ones followed magnetic tape or painted lines, or navigated by laser triangulation from reflectors mounted on walls and columns. They follow defined paths and stop when an obstacle blocks the way. AMRs, which became widespread in the 2010s, localize using natural features of the building with lidar and cameras, plan their own routes on a map, and navigate around obstacles. The distinction has blurred: many modern AGVs use natural-feature navigation, and AMRs in busy facilities are often restricted to defined lanes for predictability. Users choose based on load, path flexibility, traffic and required precision.

Fleet sizing

The basic question in any mobile robot project is how many vehicles are needed. The answer comes from the number of transport missions per hour and the average time per mission, adjusted for the share of time each vehicle is available, since vehicles charge, wait in traffic and break down.

# Monte Carlo check of an AMR fleet: queueing of random missions
import numpy as np
rng = np.random.default_rng(440)
def simulate(n_robots, missions_per_hour=48, hours=200):
    t = np.cumsum(rng.exponential(3600 / missions_per_hour, int(missions_per_hour * hours * 1.1)))
    t = t[t < hours * 3600]
    free = np.zeros(n_robots)
    waits = []
    for arrival in t:
        k = np.argmin(free)
        start = max(arrival, free[k])
        waits.append(start - arrival)
        dur = rng.normal(360, 40)                       # mission time, s
        if rng.random() < 0.15:                         # some missions follow a charge or fault
            dur += rng.uniform(300, 900)
        free[k] = start + dur
    w = np.array(waits)
    return w.mean(), np.percentile(w, 95)
for n in (6, 7, 8, 9):
    mean_w, p95 = simulate(n)
    print(f"{n} AMRs: mean wait {mean_w:6.0f} s, 95th percentile wait {p95:6.0f} s")

6 AMRs: mean wait   2316 s, 95th percentile wait   4984 s
7 AMRs: mean wait    228 s, 95th percentile wait    885 s
8 AMRs: mean wait     52 s, 95th percentile wait    278 s
9 AMRs: mean wait     21 s, 95th percentile wait    145 s

The simulation illustrates a general lesson from queueing theory: as utilization approaches 100 per cent, waiting times grow sharply, and the 95th-percentile wait, which determines whether the assembly line ever starves, grows much faster than the average. Fleet management software vendors typically offer simulation tools for this purpose, and integrators validate them during commissioning.

Goods-to-person and storage systems

Goods-to-person systems bring storage containers to stationary pickers at ergonomic workstations, often with put-to-light displays that show which order container receives each item. Removing walking can raise pick rates from perhaps 60 to 150 lines per hour in manual picking to several hundred lines per hour per station, depending on order profiles. Shelf-lifting robots drive under mobile shelving units and carry them to stations; cube storage systems stack bins in a grid with robots on top that dig out the requested bin; shuttle systems run in each level of a racking structure with lifts at the end of each aisle.

Unit-load AS/RS use cranes running in aisles to store pallets in tall racking, sometimes 30 m or more high, saving floor space and land. A crane's travel time to a storage location depends on horizontal and vertical distance; because the horizontal and vertical drives move simultaneously, the travel time is the larger of the two, the Chebyshev metric.

# Expected single-command cycle time over random rack locations (Chebyshev travel)
import numpy as np
rng = np.random.default_rng(440)
L, H, vx, vy, t_handle = 90.0, 24.0, 3.0, 0.8, 15.0
x = rng.uniform(0, L, 100_000); y = rng.uniform(0, H, 100_000)
one_way = np.maximum(x / vx, y / vy)
cycle = 2 * one_way + 2 * t_handle
print(f"mean cycle {cycle.mean():.1f} s -> {3600 / cycle.mean():.0f} cycles per hour")
print(f"rack 'shape factor' (time to end / time to top): {(L/vx)/(H/vy):.2f}")

mean cycle 70.0 s -> 51 cycles per hour
rack 'shape factor' (time to end / time to top): 1.00

Fleet management and integration

A fleet manager assigns missions to vehicles, plans routes, manages traffic at intersections and narrow aisles, schedules charging and handles exceptions such as blocked paths. Task allocation can be as simple as the nearest available robot, or it can optimize across many robots using methods related to the assignment problem. Traffic management prevents deadlocks, in which robots wait for each other in a cycle, often by reserving path segments in advance.

Above the fleet manager sit the warehouse management system (WMS), which manages inventory and orders, and often a warehouse execution system (WES) or warehouse control system (WCS) that orchestrates automation equipment in real time. Mobile robots also interact with building systems: automatic doors, lifts, fire alarm systems that must clear robots from escape routes, and conveyors where they hand over loads.

Mixing fleets from different suppliers has historically meant running several fleet managers that cannot coordinate. VDA 5050, an open interface specification developed by the German automotive industry association VDA together with the mechanical engineering association VDMA, defines how a master control system sends orders to AGVs and AMRs and receives their state, using MQTT and JSON messages. It is increasingly supported by vendors worldwide. The MassRobotics interoperability standard from the United States addresses a related goal of sharing position and status between fleets. Interoperability remains an evolving area in 2026.

Batteries and charging

Mobile robots run on batteries, now mostly lithium-ion chemistries such as lithium iron phosphate, which tolerate frequent partial charging. Charging strategies include opportunity charging during short idle periods, which keeps the fleet working with fewer spare vehicles, and scheduled charging in quiet periods. Battery capacity, charge rate and duty cycle determine the availability figure used in fleet sizing.

Safety standards for industrial mobile robots

Mobile robots raise hazards different from fixed arms: collisions with pedestrians and forklifts, crushing between a vehicle and fixed objects, loads falling from vehicles, and arms mounted on mobile bases, called mobile manipulators. Internationally, ISO 3691-4:2023 covers driverless industrial trucks and their systems, and was under revision at the time of writing. In North America, ANSI/A3 R15.08 addresses industrial mobile robots: Part 1, published in 2020, sets requirements for manufacturers; Part 2, published in 2023, covers integration of IMR systems and applications; and Part 3, for users, was in development. Canada had no national standard specific to industrial mobile robots that we could identify at the time of writing, so Canadian integrators commonly apply ISO 3691-4 and the R15.08 series, along with CSA Z434:26 for robot arms on mobile bases and provincial occupational health and safety requirements. Key design elements include safety-rated laser scanners with protective fields that grow with speed, defined operating zones, speed limits in pedestrian areas, warning lights and sounds, and site risk assessments that consider floor condition, aisle widths and mixed traffic.

Economics and operations

Warehouse automation projects are judged by throughput, accuracy, labour savings, space savings and flexibility. Highly automated systems such as AS/RS and cube storage give the highest density and throughput but are hard to move or expand. AMR systems can often be added in phases, leased, or offered as robots-as-a-service with monthly fees, which lowers the up-front investment. Peak seasons, such as the period before the end-of-year holidays, matter: a system sized for average demand may fail at the peak, while one sized for the peak sits idle much of the year, which is why many operations combine automation for the base load with temporary labour for peaks. Lesson 11 covers the financial analysis in detail.