Nico Peng
← Back to the Physical AI index

Physical AI · Family 2 · Managed-site route

In-facility mobile robots

How transport robots, loads, charging and fleet software become a working material flow.

AS OF 2026-09-02DATED · NON-EXHAUSTIVE

Supply badges: “not assessed” means this report has not established a node-wide supply rating. It does not mean supply is plentiful; product maturity and local constraints are discussed separately.

Family 2 · Managed-site route

In-facility mobile robots

AMRs, guided vehicles and autonomous lift trucks move goods through warehouses and factories. The useful unit is a working material flow: robots, loads, charging, fleet software and the connections to people and other machines.

Moving goods without fixing every movement in place

A warehouse robot may carry a shelf to a picker, follow a picking workflow, tow a cart or lift a pallet. This page covers those transport machines inside managed sites. Autonomous mobile robots (AMRs), automated guided vehicles (AGVs) and autonomous lift trucks use different navigation and handling designs, but share the need to move the right load to the right place safely and repeatedly.

They can automate transport without rebuilding every route as a conveyor. That flexibility is useful where layouts, order mixes or production flows change. Some systems follow tightly specified paths; others plan around obstacles. Neither the AMR label nor the absence of a driver tells us how much of the software is learned.

DHL's June 2024 disclosure reported 500 million picks supported by Locus robots, with deployments at more than 35 DHL-managed sites. That is repeated operational use, rather than a laboratory demonstration. People still participate in picking: the robots remove travel and coordinate the flow rather than necessarily grasping every item themselves. DHL's deployment record →

This is one route within mobile service and logistics robots. The public-space route → follows related enabling capabilities into a different job: operating around untrained people and completing deliveries or services. Keeping two diagrams avoids pretending those downstream operations are interchangeable.

AI can improve a task or coordinate an entire fleet

A concrete example · pallet handling

Finding the pallet is part of the job

Driving to a pickup point is only the start. A pallet may be displaced or wrapped differently; the vehicle must find its openings and align its forks. MiR's March 2024 pallet-jack announcement describes AI-based 3D detection using NVIDIA Jetson AGX Orin. This is a specific role for learned perception inside an otherwise engineered mobile machine. MiR's product announcement →

The finished product also needs load capacity, protective sensing, reliable stopping and charging. The AI feature changes one difficulty in the workflow; it does not replace those inputs. Pallet-jack system →

There is also intelligence above the individual robot. Amazon's DeepFleet research models interactions among large numbers of mobile robots from fleet movement data. Better coordination can reduce waiting and congestion even when the hardware is unchanged. This creates a software and data opportunity at fleet level, not just demand for a more powerful computer on every vehicle. DeepFleet research, August 2025 →

The economic question is whether these improvements reduce the effort to deploy another route or increase useful throughput in an existing site. Faster empty travel is of limited value if loading, human picking, charging or a congested handover point remains the constraint.

The enabling supply chain

Build the vehicle, then integrate the flow

The left route builds a purpose-designed robot; the upper-right route adapts an industrial truck. Both need navigation and control. Equipment joins fleet software and site engineering before it becomes an accepted working system.

Product or componentEngineering or productionSupplied outputSupply / processAlternative routeApplication-dependentDevelopment supportMaintenance / support
Each bubble opens its explanation below
Build the vehicle, then integrate the flowThe left route builds a purpose-designed robot; the upper-right route adapts an industrial truck. Both need navigation and control. Equipment joins fleet software and site engineering before it becomes an accepted working system.

Factory-built mobile robots and converted industrial trucks are alternatives, not consecutive stages. Dashed development links identify learned capabilities where used. Charging hardware enters the vehicle design; charger placement and scheduling are also part of site deployment. Ownership and service contracts are explained below.

Key players are selected examples, linked to their relevant products or work—not a market-share ranking. Some activities are performed inside a manufacturer or operator rather than sold separately. “(private)” marks unlisted businesses; listed parent groups are identified separately.

The vehicle and its load

Transport begins with a mechanical and power design suited to the task. A tote carrier, pallet jack and converted lift truck should not be treated as interchangeable equipment.

Economic nodes

Wheel drives & brakes

Supply not assessed

Motors, gearboxes, wheels, steering, brakes and encoders provide the motion system. Payload, floor friction, stopping distance and duty cycle determine the design. A low transport deck and a lift truck have very different loading requirements even when they navigate with related software.

Key players

  • Bosch Rexroth (private) — ROKIT integrated wheel drives.
  • maxon (private) — AGV and AMR wheel-drive systems.

Worth understanding

The drive system must expose the measurements and stopping functions required by the protective controller. SICK's mobile-robot guide illustrates that interface between motion and protection. Mobile-robot engineering guide →

↑ Back to the map

Robot chassis

Supply not assessed

The chassis carries the load, battery, drive units and electronics. A robot built to move shelves, carts or totes has a different body and load interface from a conventional forklift. This node is the physical platform, before its autonomous behaviour and working application are integrated.

Key players

  • MiR (private) · Teradyne — native mobile-robot platforms.
  • Locus Robotics (private) — mobile platforms for warehouse transport.

Worth understanding

MiR's range illustrates purpose-designed transport platforms. The chassis choice constrains turning space, centre of gravity and what equipment can be mounted on top. Mobile platform examples →

↑ Back to the map

Battery & charging

Supply not assessed

The pack, battery management and charging interface determine when the robot can work and how it returns to service. Opportunity charging can use idle intervals; contact or inductive charging requires compatible hardware on the vehicle and at the site.

Key players

  • Wiferion (private) — wireless charging for mobile robots.
  • Delta-Q (private) — battery chargers for mobile robotics.

Worth understanding

Wiferion supplies charging systems for mobile robots. The separate site problem is where chargers go and how a fleet shares them without disrupting material flow. Charging products →

↑ Back to the map

Load handling

Supply not assessed

Lift mechanisms, roller tops, tow couplings, forks and fixtures connect transport to the actual load. A robot that arrives at the destination has not completed its job if the pallet cannot be picked up or the receiving machine cannot accept it.

Key players

  • MiR (private) · Teradyne — pallet-handling mechanisms on its pallet-jack robots.
  • BALYO (private) · SoftBank subsidiary — autonomous reach, stacker and turret-truck configurations.

Worth understanding

MiR's pallet-jack product is one integrated example. Other tasks use different modules, and manual loading can remain part of the workflow. Load-handling example →

↑ Back to the map

Industrial truck

Supply not assessed

An existing industrial truck supplies the mast, forks, steering, brakes and load-rated structure for a conversion. Its control interfaces and condition constrain what can be automated; compatibility is a property of a named vehicle configuration, not of forklifts in general.

Key players

Worth understanding

Balyo's truck automation illustrates this alternative to a purpose-built mobile base. A retrofit does not remove the engineering needed to command and stop the host machine. Industrial-truck integration →

↑ Back to the map

Navigation and production

Sensing and software become useful only when integrated with the machine. The production route can start with a new mobile platform or a compatible industrial truck.

Economic nodes

Localization & sensing

Supply not assessed

Localization estimates where the robot is; scene and load sensing tell it what surrounds it. Protective sensing detects hazards and supports defined safe stops. These functions can share measurements without being the same function.

Key players

  • SICK (private) — lidar localization products for mobile platforms.
  • Bosch Rexroth (private) — ROKIT localization and navigation components.

Worth understanding

SICK's LiDAR-LOC uses map contours for localization and can use data from compatible safety scanners. That is a concrete reusable product, not proof that every navigation stack or safety function uses the same sensors. Localization product →

↑ Back to the map

Compute & control

Supply not assessed

Onboard compute runs navigation and, where used, learned perception. Motion and safety controllers translate commands into predictable movement and stopping. Combining them in a product does not eliminate their different timing and assurance requirements.

Key players

  • NVIDIA — Jetson compute used in MiR’s pallet-handling robot.
  • MiR (private) · Teradyne — integration of computing with robot controls.

Worth understanding

The MiR pallet-jack announcement identifies an AI compute module, while SICK describes protective controllers and scanners. These are different roles inside the same machine. AI compute example → · Protective control →

↑ Back to the map

Task & fleet data

Supply not assessed

Images of loads, route observations, intervention logs and fleet movements provide examples for developing or improving behaviour. The useful data depends on the task: pallet recognition and traffic coordination do not use the same labels or objective.

Key players

  • Amazon Robotics · Amazon — internal fleet data and DeepFleet model development.
  • Viam (private) — robot-data collection and management infrastructure.

Worth understanding

DeepFleet trains on robot positions, goals and interactions. Data creation may be captive to a large fleet owner; it still performs an economically distinct role in the development chain. Fleet-data research →

↑ Back to the map

Navigation software

Supply not assessed

Software estimates position, chooses paths, handles obstacles and coordinates the motion required to perform the task. Learned perception or planning can be added where it earns a place; an industrial vehicle need not use an end-to-end foundation model to be useful.

Key players

  • SICK (private) — localization software for mobile platforms.
  • BALYO (private) · SoftBank subsidiary — navigation software for autonomous industrial trucks.
  • MiR (private) · Teradyne — navigation integrated into its native robots.

Worth understanding

SICK offers a localization component; MiR's pallet recognition is a narrower adaptive capability. The platform developer combines such capabilities with its own navigation and exception logic. Localization interface → · Adaptive load handling →

↑ Back to the map

Robot manufacture

Supply not assessed

The native build combines the chassis, power, handling and autonomy systems into repeatable equipment. Assembly, software installation, calibration and test check that the produced unit behaves as the design requires.

Key players

  • MiR (private) · Teradyne — native AMR and pallet-jack products.
  • Locus Robotics (private) — warehouse robot platforms.

Worth understanding

This can be performed by the robot producer or manufacturing partners. It is distinct from site commissioning: the factory does not determine every route, load handover and traffic rule at the customer's facility. A finished robot offering →

↑ Back to the map

Truck conversion

Supply not assessed

Conversion fits autonomy hardware and software to the compatible truck, integrates steering and braking, calibrates sensors and tests the modified vehicle. The customer can preserve a familiar vehicle architecture while adding a different operating capability.

Key players

  • BALYO (private) · SoftBank subsidiary — robotization of Linde and Hyster-Yale truck platforms.

Worth understanding

It is an alternative production route, not an extra step required after manufacturing an AMR. Balyo's offering grounds the route without establishing that conversion dominates the market. Conversion offering →

↑ Back to the map

Mobile-robot equipment

Supply not assessed

The output is a functioning transport robot or autonomous truck with specified load, navigation and safety capabilities. Equipment can be sold, leased or retained by the robot provider. It still needs an application and working connections at the site.

Key players

  • MiR (private) · Teradyne — AMRs and autonomous pallet jacks.
  • Locus Robotics (private) — warehouse robots supplied through its service offering.
  • BALYO (private) · SoftBank subsidiary — autonomous industrial-truck equipment.

Worth understanding

MiR's product specification describes the vehicle; Locus describes a service-based route for supplying robot use. Both must eventually support an actual material-handling workflow. Equipment → · Service model →

↑ Back to the map

Making a site work

Robot equipment is one input to a coordinated material flow. This downstream work explains why fleet software, integration and continuing support belong in the supply chain.

Economic nodes

Fleet orchestration

Supply not assessed

Fleet software assigns jobs, coordinates traffic, monitors units and schedules charging. It connects individual robots to shared work and constraints. Multiple robots in a corridor can create congestion that no improvement to one robot's perception solves.

Key players

  • MiR (private) · Teradyne — fleet management software.
  • InOrbit (private) — multi-robot orchestration.
  • Amazon Robotics · Amazon — internally operated fleet intelligence.

Worth understanding

MiR Fleet is a supplied example. Amazon's DeepFleet research shows a learned approach to coordination, but a fleet-management product and a research model are not automatically substitutes. MiR software → · Coordination research →

↑ Back to the map

Site & workflow integration

Supply not assessed

The deployment team maps routes, defines pickup and drop-off rules, places chargers and connects doors, conveyors and warehouse or manufacturing systems. It also determines what people do when a load is missing or a route is blocked.

Key players

  • Locus Robotics (private) — deployment into customer warehouse workflows.
  • MiR (private) · Teradyne — fleet and enterprise-software interfaces.

Worth understanding

DHL's repeated Locus deployments illustrate why this activity matters: useful scale is a repeatable working process across facilities, not simply a larger delivery of vehicles. Multi-site deployment →

↑ Back to the map

Validation & commissioning

Supply not assessed

Commissioning verifies the actual operating area, traffic interactions, safe stops, staff procedures and acceptance criteria. It turns a configured system into one the site can use under defined conditions.

Key players

  • Pilz (private) — on-site application risk assessment, validation and commissioning support.

Worth understanding

ISO 3691-4 covers driverless industrial trucks and their systems. The standard is a requirement, not a supplier node; the engineering and testing used to meet the applicable requirements are the activity shown here. Industrial-truck standard →

↑ Back to the map

Working site fleet

Supply not assessed

The accepted output is a fleet performing the site's material flow. Its useful throughput depends on loading, unloading, order release, charging and human collaboration as well as robot movement.

Key players

  • Locus Robotics (private) — deployed warehouse fleets, including DHL sites.
  • BALYO (private) · SoftBank subsidiary — installed autonomous pallet-movement systems.

Worth understanding

DHL's operational pick count is a stronger sign of use than a shipment target, but it does not disclose a comparable cost per pick for every site. The system should be judged against the workflow it replaces. Operational evidence →

↑ Back to the map

Fleet support

Supply not assessed

Service maintains batteries, wheels, sensors and software, diagnoses faults and restores robots to use. The fleet also needs someone to resolve exceptions and reconfigure work when the facility changes.

Key players

  • Locus Robotics (private) — maintenance and updates within its RaaS offering.

Worth understanding

Under a robots-as-a-service arrangement, much of this can be bundled with access to the robots. A recurring bill alone does not prove that the supplier has taken responsibility for every customer throughput outcome. Locus service description →

↑ Back to the map

Throughput depends on the whole workflow

A mobile robot can remove walking or driving time while people continue loading, picking or handling exceptions. The benefit depends on which activity was limiting output. More robots can increase capacity, but can also create queues at chargers, intersections and receiving stations.

This makes integration repeatability an important economic test. A supplier that can commission the next site quickly, connect existing systems and support a mixed operation offers more than a vehicle bill of materials. Conversely, a deployment that needs extensive re-engineering at every site may remain service-intensive even when the robots themselves are standardized.

Equipment sales and service subscriptions allocate capital and maintenance differently. Evaluate useful operating hours, exception labour, charging, spares and site changes under either model. The publicly disclosed deployments establish demand for working systems; they do not establish one industry-wide margin or payback period. Customer deployment → · Service structure →