Physical AI · Family 5
Driverless haulage, automated field tasks and precision implements share inputs without doing the same job.
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.
Agricultural, mining and construction machines use AI and automation to drive, handle material or control a working tool. Rugged hardware and machine interfaces recur, but autonomous haulage, tractor fieldwork and precision implements have different supply and deployment routes.
A haul truck moving around a mine, a tractor tilling a field and a sprayer selecting individual weeds all use physical automation. They do not remove the same work. The truck can operate without a cab driver inside a managed area. A tractor may automate one field operation but still need people for transport and setup. A smart implement can improve a task while the driver remains in the seat.
They belong together provisionally because rugged sensing, positioning, control electronics, machine interfaces and field service recur. But the map must preserve their different outputs. A farming kit is not simply a smaller mining-autonomy system, and a successful machine-control product need not progress toward a fully driverless farm.
The starting asset often already exists. This makes compatibility, conversion and dealer installation important alongside new-machine manufacture. Electrification is a separate design choice: autonomy can be integrated with diesel, diesel-electric or battery-powered equipment. There is no universal battery dependency across this family.
Autonomous mining did not begin with the recent AI boom. Komatsu dates commercial FrontRunner deployment to 2008 and reported its thousandth commissioned ultra-class autonomous truck in April 2026. That is a substantial operating base, but it does not establish that general learned perception now controls every mine task. Commissioned haulage fleet →
The newer opportunity includes narrower tasks made practical by better perception and control. Deere reported that See & Spray covered more than five million acres in the 2025 growing season, with average non-residual herbicide savings of nearly 50%. Those are the supplier’s reported results for selective spraying—not five million acres of driverless tractor operation. Selective-spraying evidence →
AGCO’s March 2026 account describes OutRun as a commercially available retrofit route for specified tasks, including tillage. The important development is a usable capability that can be added to compatible existing equipment. Its goal of broader farm autonomy by 2030 is separate from what the current kit performs. Task-specific retrofit →
The enabling supply chain
The upper inputs feed a machine-autonomy kit and, where relevant, a precision implement. The machine route can use factory fit or conversion. An equipped machine and a commissioned site fleet are distinct outputs; implement-level automation is shown separately on the left.
This is a family map, not one universal machine architecture. Amber links identify application-dependent inputs or deployments. The implement branch can retain a human driver. Mining, fieldwork and construction have different assurance requirements, workloads and support arrangements.
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.
These inputs recur across field machines, but their specifications and importance depend on the task. A shared node does not imply identical hardware.
Positioning combines available satellite, inertial and local references to estimate machine location. The accuracy and availability needed for a haul road, a crop row and a grading task differ. Underground or obstructed environments can require a different mix from an open field.
Hexagon’s positioning portfolio illustrates a separately supplied enabling market. Positioning products →
↑ Back to the mapCorrection services improve suitable satellite-positioning solutions. They can be a recurring input rather than part of the machine purchase. Availability and precision still depend on the receiver, environment and chosen service; not every autonomous system relies on this route. Trimble correction services →
Cameras, radar, lidar and task sensors observe obstacles, terrain, crops or the working area. Dust, vibration, mud and changing light affect their installation and performance. Detecting a weed for a nozzle is a different job from detecting another vehicle on a haul road.
Embedded computers process observations and issue machine or implement commands. Power electronics and controllers must work under the host equipment’s environmental and timing requirements. Rugged packaging, interfaces and support can matter as much as peak AI performance.
Field images, machine behaviour, interventions and test scenarios support development and validation. Data needs are task-specific: crop recognition, soil work and haul-road interactions cannot be treated as one interchangeable training set. Captive collection by the machine producer still performs a distinct development function.
Electro-hydraulic, steering, braking and implement interfaces translate commands into physical action. Compatibility includes how the machine behaves when commands or communications fail. A software connection alone does not establish a safe conversion.
Danfoss offers electro-hydraulic steering components for auto-guided systems, illustrating the interface market. Steering interface →
↑ Back to the mapCapability engineering and kit production meet a compatible base machine. Factory fit and conversion provide different paths to an equipped asset.
Engineering combines perception, positioning, control and fault handling for a particular machine and task. It defines supported operations and compatible equipment. The resulting capability may automate motion, a working tool, or both; these are choices to specify rather than levels to assume.
Manufacture packages sensors, computing, wiring and interfaces into an installable configuration, then checks the produced kit. Standardizing that work can make repeated installations easier. It does not eliminate the need to validate the host machine or its intended task.
The supplied kit adds a bounded operating capability to compatible equipment. It may be sold with software access, installation or support. OutRun is an example of a retrofit offering; its applicability should be read at the supported task and tractor level. Retrofit offering →
The tractor, haul truck or construction machine supplies the load-bearing structure, propulsion and working capacity. Existing fleet life can make conversion attractive, while a new vehicle can be designed for automation from the outset. Neither route makes different machine classes interchangeable.
Factory fit integrates the automation into a new or purpose-configured machine. This lets the equipment maker coordinate interfaces, production and support. Komatsu’s autonomous haul-truck offering is one example of equipment and autonomy being supplied within a coordinated product route. Integrated haulage equipment →
Conversion installs the kit on a compatible machine, calibrates sensors and controllers, and tests the modified configuration. It is a production alternative to factory fit, not an obligatory later upgrade.
Epiroc’s Roy Hill account describes a heterogeneous fleet using Caterpillar and Hitachi trucks, showing that an autonomy integration can span multiple equipment makers. Mixed-fleet integration →
↑ Back to the mapValidation checks the integrated equipment, intended task, operating limits and fault responses. The work depends on the machine and environment. A mining-system assurance case is not automatically valid for a tractor or a machine working around a construction crew.
The output is a machine with a specified automated capability. An owner can put a task-specific machine to work within its supported conditions; some applications additionally need a coordinated site system. This node does not imply unattended operation across every part of an owner’s day.
A coordinated site fleet and a precision implement deliver different benefits. Their separate endpoints prevent task automation from being misread as universal driverlessness.
A coordinated site can require communications, a working map, traffic rules and software that dispatches multiple machines. Those are supplied systems, unlike a general label such as “site readiness.” Their role is especially visible in autonomous haulage.
Epiroc describes the virtual mine map and connection with Roy Hill’s remote operations centre. Site-system example →
↑ Back to the mapSite engineering configures the equipment, routes, traffic interactions, communications and operating procedures together. Training and acceptance checks establish how the system will be used. This can be substantial work rather than a software activation at the end of a truck delivery.
Caterpillar’s quarry account describes adapting a mining-autonomy system for a different scale of operation. Quarry deployment route →
↑ Back to the mapThe customer receives a coordinated working fleet for a defined site operation. Its output depends on loading, dumping, maintenance and interactions with other equipment. A driverless truck count alone does not measure the productivity of that whole system.
The working tool combines task sensing and control with an actuator such as a spray nozzle or grading mechanism. It can create value through input savings or work quality while a person continues driving. Deere’s selective spraying is a concrete example, not a proxy for whole-machine autonomy. AI at the working tool →
The implement is fitted to its host, connected to controls and calibrated for the task. Nozzle timing, tool position and operating speed must agree. A good perception result becomes useful only when the action is delivered at the right place and time.
This output can improve chemical use, accuracy, quality or rework without eliminating the machine operator. Its economic comparison should therefore use the inputs saved and work improved, rather than assume a full wage has been removed.
Installation support, calibration, repairs, spare parts and software service keep equipment available through demanding duty cycles and short work windows. A distant development team cannot substitute for every field repair. The dealer or service partner can therefore remain important even when software is the new capability.
For a haulage operation, the case can involve utilization, predictable flow and reduced exposure of people to hazardous work. For a field task it can involve timeliness and labour flexibility. For a precision implement it may be reduced inputs, better quality or less rework. These mechanisms are not interchangeable, and a single “cost per robot” comparison hides them.
Physical work also remains around the automated task: moving between fields, refuelling, loading, clearing faults, changing implements and maintaining the site. A machine’s autonomous hours are not the same as an owner’s labour hours saved.
The family is worth keeping together for now because the enabling supplier roles recur. Its internal routes should remain visible, and a route deserves a separate map if its actual interfaces become too different to explain here. No market-size threshold or industry label alone decides that boundary.