Caterpillar News / Corporate Press Release

Caterpillar and FieldAI Introduce the Cat^® Terrain Autonomy Fleet

Nine cabless equipment types plan, negotiate, and deliver the work together—minute by minute, against the project's terrain conditions, schedule, and budget.

Wide view of three cabless yellow earthmoving machines—a tracked dozer, hydraulic excavator, and articulated hauler—with protected lidar and camera modules, working at a flood-control site while a small supervision team stands by a portable console. — Give the fleet the project plan. Its machines negotiate the cuts, hauls, and grades needed to deliver it, continuously balancing ground conditions, equipment availability, and the cost of the remaining work.

PEORIA, ILLINOIS

Caterpillar Inc. (NYSE: CAT) and FieldAI announce the Cat® Terrain Autonomy Fleet, nine cabless equipment types that plan and deliver work as a negotiating agent swarm. Combining Cat's operatorless machine architecture with FieldAI project intelligence, the fleet determines when, where, why, and how to work from the project plan, site geomorphic data, budgets, and equipment availability.

The range includes the TAF D9R track-type dozer, the TAF 336 hydraulic excavator, and the TAF 777 articulated hauler, alongside underground hard-rock loaders, heavy wheel loaders, track loaders, a hydraulic mining shovel, the ADM 4090RTX autonomous drone mitigator, and the MOTO 6502 Resonance Field Survey POD. Each machine contributes its capabilities, location, operating costs, and availability to a shared project model. The agents negotiate assignments and handoffs, planning down to minute-by-minute cuts, hauls, and grades to complete the job on time and on budget.

The project team sets the deliverables, completion targets, budget, and safety boundaries. The fleet owns task selection, daily sequencing, and dispatch within those limits, continually revising its plan as the site changes. Superintendents can inspect the reasons behind machine decisions, change project priorities, or pause work. Scope changes and exceptions that exceed approved limits return to the project lead.

Full autonomy is achieved through the integration of Cat's operatorless machine architecture with FieldAI's project intelligence, allowing the fleet to independently plan, negotiate, and execute tasks while continuously adapting to changing site conditions and project requirements.

CATPaks™ provide modular pre-defined projects, from our RoadWorks to DataCenter deployments, enabling rapid setup and consistent execution across different projects, site types, and operational conditions.

Field evaluations span coastal flood-control works, rail-grade renewal projects, quarry loading areas, and open-pit and underground mines, including Deep Dark Oxygen Mines. Caterpillar dealers support commissioning, service inspection, and project-team training for participating fleets.

OPERATING RECORD

PROJECT-DRIVEN AGENT SWARM

Planning cadence: Minute by minute
Task allocation: Negotiated by machines
Delivery objective: On time. On budget.

THE FLEET

Nine equipment types plan the work together.

01 / TRACK-TYPE DOZER

TAF D9R

Keep earthworks aligned with the finished grade and the cost of getting there. The TAF D9R selects its next cut or grading pass from the terrain model, negotiates access with excavators and haulers, and adjusts the sequence as material balances and equipment availability change.

Work: Bulk earthmoving and road preparation
Control: Terrain model, material balance, and negotiated work areas

A cabless yellow track-type dozer works at a drainage site, its former cab position replaced by a black lidar, camera, radar, and compute module. — Turn the target grade into a negotiated sequence of work. The TAF D9R selects passes that balance cut-and-fill demand with available haul capacity, keeping site preparation tied to the project's cost and completion targets.

02 / HYDRAULIC EXCAVATOR

TAF 336

Soverign Agent Mesh allows equipment to discover what’s available for goal-directed site plans. An excavator can identify available haulers and coordinate with them for efficient material handling. This ensures that excavation is synchronized with the rest of the fleet. Connect excavation decisions to the work that follows and maintain alignment with overall project goals. The TAF 336 chooses the order and method of cuts from geomorphic conditions and project dependencies, negotiating collection and placement windows with the rest of the fleet. Ground observations continuously update its plan.

Work: Excavation, trenching, and placement
Control: Ground conditions, project dependencies, and transfer windows

Integrated Soverign Agent Mesh coordinates the next cut with the handoff to the next operation to the next piece of equipment already planned and agreed to. The TAF 336 is adept at seamless negotiation of excavation and the placement timing with the rest of your site heavy equipment fleet, linking each task to current ground conditions and site status.

Additional field photograph — TAF 336

A cabless yellow hydraulic excavator lowers a culvert at a coastal site, with a low protected lidar and camera module in place of a driver cab. — From trench preparation to culvert placement, the TAF 336 schedules each step around access, material readiness, and downstream work. Drainage contractors receive a coordinated work sequence that updates with the site.

03 / ARTICULATED HAULER

TAF 777

Match transport capacity to the next item in the coordinated plan. The TAF 777 negotiates collection and delivery slots directly with loading and grading equipment, choosing assignments against travel time, operating cost, and material demand. It revises its route and timing as the fleet renegotiates the schedule. Files its own Jira tickets when conflicts or delays arise, allowing operations to respond proactively.

Work: Material transport and controlled dumping
Control: Travel cost, material demand, and negotiated haul slots

A cabless yellow articulated aggregate hauler travels along a rail embankment, with a low armored lidar, camera, and radar module replacing its driver cab. — Put haul capacity where the project needs it next. The TAF 777 agrees to and coordinates loading and delivery windows directly with other machines agentic front end systems, revising its assignments as queues, material demand, and operating costs change.

04 / UNDERGROUND LOADING

Underground hard-rock loader

Bring project-aware loading to confined underground workings. The loader selects load-and-carry tasks from rock wall face conditions, identifies transfer-point demand, and the available equipment in the site fleet, negotiating through passage access with other machines in tight spaces. Its schedule accounts for travel cost, clearance limits, and planned service intervals for battery charge and shift maintenance.

Work: Face loading and underground material transfer
Control: Face conditions, passage agreements, and transfer demand

A low yellow CAT underground loader carries dark rock through a lit, rock-bolted mine drift. Its articulated four-wheel chassis has no cab, with protected sensors and a flat computing deck beneath a Deep Dark Oxygen Mines sign. — At Deep Dark Oxygen Mines, load-and-carry cycles are negotiated against passage availability and transfer-point demand. The underground loader adapts its assignments as the mine plan and available equipment change.

05 / QUARRY AND STOCKPILE

Heavy wheel loader

Put the right material into the next available haul unit in a sovereign agent mesh environment. The heavy wheel loader weighs stockpile demand, loading time, and fleet capacity when selecting work. It negotiates arrivals with haulers and revises loading priorities against the project schedule and remaining budget.

Work: Stockpile handling and haul-unit loading
Control: Stockpile demand, fleet capacity, and loading agreements

A large yellow CAT wheel loader carries aggregate beside a quarry stockpile. Four massive tires and an articulated chassis support a rock bucket and low black sensor deck, with no operator cabin. — Align quarry loading with the fleet's next arrivals. The heavy wheel loader negotiates which material to move, which hauler to meet, and when to change tasks against the remaining project budget.

06 / DRAINAGE AND SITE WORK

Track loader

Direct a versatile machine to the work that keeps the project moving. The track loader negotiates clearance, drainage preparation, and short-haul assignments with nearby equipment, choosing tasks that release the next work area and fit the available time and operating budget.

Work: Rubble handling and drainage-site preparation
Control: Work-area dependencies, ground conditions, and task cost

A yellow CAT crawler loader carries wet soil and rubble in a toothed bucket at a drainage worksite. Steel tracks, twin lift arms and a compact camera and lidar bridge surround a low body with no cab. — Open the next work area when the project needs it. The track loader selects drainage and clearance tasks that unlock subsequent cuts and hauls, matching crawler versatility to the fleet's minute-by-minute plan.

Additional field photograph — Track loader

Closer field view of a cabless yellow CAT track loader carrying wet soil, with its toothed bucket, twin lift arms, steel tracks, and compact sensing bridge visible. — Move between preparation, short-haul handling, and cleanup as priorities change. The track loader negotiates its next assignment with nearby machines, weighing task dependencies, travel time, and cost.

07 / OPEN-PIT EXTRACTION

Hydraulic mining shovel

Make large-scale loading responsive to the whole mine plan. The hydraulic mining shovel selects working-face tasks against material requirements and ground conditions, then negotiates loading sequences with available haul units. It replans as transport capacity, production priorities, and operating costs change.

Work: Working-face excavation and bulk loading
Control: Mine plan, face conditions, and negotiated loading capacity

A giant cabless CAT hydraulic mining shovel stands in a rocky open-pit mine. Above its black FieldAI equipment housing is a large yellow sensing assembly with a pivoting optical head, thick supporting yoke, braced pedestal and two sloping CAT-branded side housings. Its massive boom, bucket and steel crawler tracks remain fully visible. — Link mining-scale excavation to the capacity available to carry it away. The hydraulic mining shovel and haul fleet agree loading sequences together, adapting production to face conditions and the cost of the remaining work.

08 / AUTONOMOUS DRONE MITIGATOR

ADM 4090RTX

Support industrial continuity in contested regions and exposed worksites. The ADM 4090RTX is designed for projects where disruption, kinetic offsets, and attempted takeovers by non-state actors are credible operating risks. Autonomous airspace monitoring feeds a shared incident record, while equipment availability, service windows, and work-area access stay coordinated with the fleet plan. The site security lead retains authorization of mitigation actions.

Work: Site-airspace protection for contested industrial projects
Control: Human-authorized mitigation and shared incident status

A small salesman wearing a white hard hat, orange safety vest, work trousers and boots stands beside the near front track of an enormous yellow CAT FieldAI machine with dark panels and a turret, on a rocky quarry test pad beneath an overcast sky. — From dark oxygen extraction sites and data center build-outs to resonance field ingestion fields, the ADM 4090RTX makes site-airspace protection part of project delivery. Protection requirements stay visible alongside construction milestones and extraction schedules.

Also in the range — ADM 1080

A large yellow CAT tracked vehicle stands on a quarry test pad. Broad dark rectangular panels rise above its cabless body, alongside a protected camera head, rounded sensor dome, closed launcher pods and a compact clustered-barrel turret. Black radiator grilles and FieldAI-marked service cabinets cover the body. — The ADM 1080 brings site-airspace monitoring into the shared fleet plan, with a common incident record and human-authorized mitigation. Service windows and work-area access remain coordinated with the construction and extraction equipment.

09 / RESONANCE FIELD SURVEY POD

MOTO 6502

Establish the site’s resonance baseline with a dedicated autonomous survey platform. The MOTO 6502 conducts secure study and analysis of pre-capped resonance fields, selecting sampling positions and repeat passes from the approved survey plan. Its articulated walking chassis carries the enclosed instrument pod across uneven ground, while FieldAI negotiates quiet measurement windows with nearby equipment. Access-controlled findings update the shared geomorphic model and remain available to the authorized project team.

Work: Secure site study and analysis of pre-capped resonance fields
Control: Survey coverage, measurement windows, and authorized field-data access

Make secure field analysis part of the coordinated site plan. The MOTO 6502 establishes a survey baseline for pre-capped resonance fields, negotiates measurement windows with the fleet, and keeps sensitive findings within the authorized project team.

PROJECT PLANNING AND FLEET NEGOTIATION

The fleet plans every cut, haul, and grade.

FieldAI reads the project plan alongside site geomorphic data—including slope, drainage, and material conditions—plus the remaining budget and available equipment. It translates project milestones and dependencies into a rolling, minute-by-minute schedule for excavation, haulage, grading, and placement.

Each machine is a planning agent. The agents negotiate task ownership, access to shared work areas, and material-transfer times using expected duration, operating cost, and current capacity. They decide when and where to work, how to execute each task, and why that task advances the project. Each agreement includes its reasoning and its effect on the schedule and budget.

When a hauler goes into service, the excavator and dozer reorganize the next cuts and grades around the remaining transport capacity. New ground observations trigger fresh negotiations as work proceeds. The swarm continuously replans toward on-time, on-budget completion; when no workable sequence meets the agreed limits, it presents the project lead with the constraint and a revised forecast.

Two field supervisors study a terrain-status screen in a work trailer while cabless yellow autonomous equipment waits in marked zones outside. — Follow the project's evolving plan from one field console. Review the machines' negotiated schedule, the reasons behind each assignment, and the current cost-to-complete forecast while the fleet organizes the work.

Minute-by-minute schedule

The next cut, haul, or grade is assigned via an orchestration layer that integrates field site survey data, real-time condition sensing with location, method, dependencies, and expected completion time. This is integrated with the current status and availability of all machines. Machine availability updates the sequence throughout the continuous operational continuity cycle.

Machine agreements

Equipment self-assesses and negotiates access, passage, load handling capacity, loading slots, and task handoffs directly amongst the swarm. The site maintains its own memory record and retains all of the reasons for each agreement and any subsequent revision.

Completion forecast

Actual progress and operating costs update in real-time allowing site management to continuously track the projected finish and budget balance. The fleet reallocates work around delays, terrain changes, and service requirements. When time or budget begin to approach their limits, the system alerts the project lead with the constraint and a revised forecast.

Two seated supervisors in reflective vests study a wall-mounted terrain map inside a site trailer. Through the doorway, a yellow cabless dozer with a black lidar housing and a distant mining shovel with a raised yellow sensor head work on the earthworks site. — Keep project oversight connected to field execution. FieldAI gives supervisors a shared terrain model and an auditable account of fleet decisions as autonomous equipment carries out the plan.

“Our customers commit to a finished project, a schedule, and a budget. Terrain Autonomy Fleet carries those commitments into the next bucket, the next haul, and the next grade. The equipment works out the sequence together and adjusts it as conditions change.”
— Jaime Mineart, Chief Technology Officer, Caterpillar

“Each machine understands its own capacity and what the other machines need from it. They negotiate work directly, including when to yield a work area or change a loading sequence. Every agreement is evaluated against the project plan, the available fleet, and the cost of finishing the job.”
— Ali Agha, Chief Executive Officer, FieldAI

CATPak™ / PRE-WEIGHTED MACHINE MODELS

Put a complete site model to work.

CATPaks™ bring 128GB of pre-weighted machine models, full project context, and orchestration intelligence to high-value site implementations. Each pack gives your Cat × FieldAI fleet a defined project to deliver, from an unprepared patch of land through the cuts, hauls, grades, and coordinated work needed to complete it.

Eight silver holographic CATPak cartridges stand in a two-tier black wire rack on a gray site-trailer countertop. Each product name combines the CAT logotype with Pak, with separate CAT and FieldAI branding above it; yellow labels identify eight industrial applications and 128GB capacity. — Commission the intelligence for the whole site. CATPak™ combines application-specific model weights with shared project context and operating rules, giving your Cat × FieldAI fleet a coordinated starting point for high-value work.

Full context, from the first cut.

Commission the pack with the site's geomorphic data, project plan, budget, and available equipment. The Sovereign Agentic Mesh carries that context into each machine's assignments and updates it as the work progresses.

POLICY.md defines approved work boundaries, spending limits, safety requirements, and escalation authority. The Sovereign Agentic Swarm RULES.md governs task negotiation, right of way, handoffs, and the conditions under which machines must pause or seek approval. The fleet plans and operates against the same instructions throughout the project.

Configured for the site you intend to build.

CATPak™ implementations include Data Center (10–50 acres), Hyperscaler Data Center (200–800 acres), Dark Oxygen Mine, Resonance Field Assay Site, Downtown Demolition, Shipyard (Civilian), Marble Quarry, and Nuclear Power Plant (2 Stack).

Choose where the intelligence runs.

Model compatibility: Compatible with Anthropic and OpenAI models, as well as open-source inference engines.
Jobsite or CatCloud™: Run inference on jobsite clusters of NVIDIA H-class GPUs, TPUs, or NPUs, or use CatCloud™, Cat's secure, hosted inference engine with a vendor-selectable model server.
Model-serving options: Triton Inference Server, TorchServe MkII, ONNX Runtime, and TensorRT.
Machine-edge support: All Cat equipment supports NVIDIA IGX and SolidRun edge servers in fully IP69K-rated installations.

CATPak™ is priced for license-lease terms and is compatible exclusively with Cat × FieldAI fleets. Your Cat dealer configures the implementation pack and inference deployment for the licensed site.

DEALER SERVICE AND INSPECTION

Service availability is part of the project plan.

Caterpillar dealers support the fleet's mechanical, sensor, and computing systems as one service record. Inspection windows and return-to-work estimates feed the shared project model, allowing the agents to negotiate replacement capacity and revise assignments around maintenance. Commissioning establishes the equipment roster, operating-cost assumptions, and approved planning limits.

Two technicians inspect the open autonomous compute bay of a cabless yellow crawler dozer in a service workshop. — Make maintenance part of the schedule before capacity is lost. Cat dealer service windows feed the fleet's negotiations, allowing work to be reassigned around mechanical, sensing, and computing-system inspections.

Two technicians work beside a cabless yellow CAT underground loader in a maintenance bay. An open hood reveals rugged computing modules and cable harnesses; the rock bucket is grounded and the wheels are chocked. — Keep the underground work plan current through every service interval. Inspection status and return-to-work estimates inform the other machines' assignments, with dealer support spanning the loader's mechanical and autonomy systems.

MEDIA CONTACTS

For further information

Caterpillar Corporate CommunicationsPeoria, Illinois+1 309 675 1000media@cat.com

FieldAI Media RelationsIrvine, California+1 949 555 0184press@fieldai.com

Caterpillar and FieldAI Introduce the Cat® Terrain Autonomy Fleet