The Real Story from ConExpo 2026 Isn't the Machines — It's the Architecture

By OpenAutonomy.com Editorial Team

Walk the grounds of ConExpo‑Con/Agg 2026 in Las Vegas, and the surface impression is familiar: bigger iron, cleaner drivetrains, sharper in‑cab screens. But step back from any one booth and another picture comes into focus. In the first quarter of 2026, industry heavyweights drew clear architectural lines. At CES (Jan 7), Caterpillar previewed Cat AI Assistant and an edge‑AI jobsites vision built with NVIDIA's Jetson Thor and Riva speech models. At ConExpo (Mar 3–7), Hitachi Construction Machinery previewed its LANDCROS direction and demonstrated an autonomous haulage solution (AHS) aimed at mixed‑fleet integration. Epiroc linked the Las Vegas floor to a live autonomous SmartROC D65 operating at Luck Stone's Virginia quarry via its Common Automation Panel (CAP), underscoring supervised autonomy in action. Komatsu showcased a retrofit path to quarry autonomy with Smart Quarry Autonomous, emphasizing staged adoption.

Same quarter. Same industry. Fundamentally different ideas about how autonomy should work. This isn't a show recap. It's a map of the architectural choices the mining and quarry industry is making now — choices that will shape procurement, data ownership, and operational flexibility for years.

Why Architecture, and Why Now

Autonomous haulage isn't new. Caterpillar and others have established long‑running, large‑scale deployments. Caterpillar reports autonomous trucks have logged 325+ million kilometers and moved 8.6+ billion tonnes globally. What's changed in 2026 is not just capability; it's the clarity with which multiple credible OEMs are declaring architectural positions. Those positions are not converging. ConExpo and CES crystallized the divergence.

Architecture is stickier than hardware. You can swap a truck; it's far harder to swap the control philosophy, data interfaces, dispatch logic, and integration model that connect fleets to operations. The decision mines make in 2026, whether explicit or implicit, will define upgrade paths and switching costs for the next decade. Three models are visible enough to describe.

Model 1 — The Integrated Ecosystem Approach

Core idea: Optimize vehicle, autonomy stack, intelligence layer, and fleet management as a single, tightly coupled product. Caterpillar's Q1 announcements are the clearest current expression: Cat AI Assistant (edge, in‑cab), NVIDIA Jetson Thor for on‑machine inference, and a "digital nervous system" vision that coordinates fleets and sites. The story isn't only autonomy; it's vertical integration. Machine, AI, data, and site systems designed together under a unified support model. The company's long‑running autonomy record provides proof points at scale.

In practice: A Western Australia gold mine could standardize on Cat trucks, Command for haulage, MineStar for fleet management, and in‑cab AI assistance, gaining deep integration, streamlined support, and consistent performance baselines under one service network.

The tension: Integration simplifies operations but concentrates control. Over time, data formats, workflows, and institutional knowledge become vendor‑specific. Today's optimization can become tomorrow's switching cost. The question: does system‑level optimization trade off against strategic optionality? Integrated models operate within the same broader standards landscape as open ones: ISO 17757 (autonomous machine safety), EMESRT safety frameworks, and ISO 21815 (collision avoidance) all apply regardless of architecture. The distinction is not that one model is standardized and the other is not; it's that open platforms rely on interoperability‑specific standards (like ISO 23725) to coordinate across vendors, while integrated systems manage that coordination internally.

Model 2 — The Open Platform Approach

Core idea: Decouple the autonomy stack from vehicle hardware using published interfaces, so different vendors can contribute layers of the system. At ConExpo, the LANDCROS direction and AHS demos were the most explicit statement on the show floor: a mixed‑fleet‑capable AHS (co-developed with Wenco International Mining Systems) intended to coordinate equipment across brands. The "Open" in LANDCROS is not just a slogan; it signals collaboration with external technology providers and systems built to operate alongside manned equipment during transition.

This approach treats autonomy as shared infrastructure the mine can own and control at the system level. ISO 23725:2024 provides a technical foundation here, defining the FMS–AHS interoperability layer (APIs for dispatch, map sharing, telemetry, and production coordination) that allows open systems to scale beyond a single site.

In practice: A mixed‑fleet copper operation could deploy one AHS across multiple OEM trucks, integrate it with its preferred FMS, and add vehicles or swap autonomy providers without re‑engineering the entire control layer, provided vendors conform to the same interface contract and someone is accountable for integration and safety case management.

The tension: Openness demands rigor. Multiple contributors increase integration complexity and distribute accountability. That is manageable at small scale, but coordination, validation, and safety certification across vendors become genuinely hard problems as fleets grow and autonomy levels increase. Who plays integrator‑of‑integrators and owns end‑to‑end performance? Who holds the safety case when the stack spans three or four companies? The open model's long‑term viability depends on answering those questions — not just on the elegance of the interface standard.

Model 3 — The Emerging Middle Ground (Human‑in‑the‑Loop, Graduated Autonomy)

Not every deployment fits neatly into "open" or "integrated." A third space is taking shape: less consolidated as an architecture, but increasingly visible as a design philosophy. What ties it together is not a shared technical stack but a shared premise — that autonomy works best when introduced as a graduated capability with humans still in the decision loop, rather than as a binary switch from manned to unmanned.

The implementations look different from each other, and that's part of the point.

Epiroc connected the show floor to a live autonomous SmartROC D65 drilling in Virginia, supervised via the Common Automation Panel. The machine handled task execution while humans stayed in the decision loop. This is remote‑supervised autonomy: the machine performs, the human monitors and intervenes.

Gravis Robotics demonstrated a different variant: augmented machine guidance where an operator teaches a trench routine and an audience participant triggers the excavator to complete the task autonomously using Gravis Copilot. Here the human defines intent; the machine executes.

Komatsu highlighted Smart Quarry Autonomous, a commercially available autonomy solution for select Komatsu haul trucks, deployable as a retrofit without heavy site infrastructure. This is staged adoption: a mine introduces autonomy incrementally alongside its existing manned fleet.

These are distinct operational models: remote supervision, human‑taught execution, and retrofit‑based fleet transition. They share a design center rather than an architecture. The relationship between the human and the machine is the primary variable, not the interface between vendors (as in open platforms) or the depth of system integration (as in integrated ecosystems). That diversity is why this space hasn't consolidated into a single product category the way haulage autonomy has. It may never. The appropriate human‑machine relationship likely varies by task, by operation, and by organizational readiness.

Structural Characteristics of Each Model

Note: These are tendencies of each approach; real deployments often blend elements.

Dimension	Integrated Ecosystem	Open Platform	Emerging Middle Ground
Interoperability	Primarily within the vendor's stack; selective cross‑brand support may exist	Interface‑first; aims at mixed‑fleet via FMS–AHS standards (e.g., ISO 23725)	Varies; often interoperable at the supervision/control layer (e.g., remote panels)
System optimization	Centralized within one stack; strong baseline coherence	Distributed across vendors; performance depends on integration quality	Contextual; optimizes human–machine workflows
Vendor dependency	Deeper dependency as more layers are adopted	Lower by design (standardized APIs); integrator role critical	Moderate; modular at equipment level, proprietary elements possible
Scalability model	Platform‑driven; scales with vendor roadmap	Modular; components added or replaced independently	Network‑driven; scales with connectivity and remote‑ops capacity
Data governance	Data flows through vendor platforms; portability varies by contract	Mine retains access via open data standards and contracts	Shared; governance tied to connectivity architecture
Upgrade path	Vendor‑managed, system‑wide coordination	Component‑level upgrades; re‑integration may be required	Graduated; autonomy level can increase over time
Safety standards framework	ISO 17757, EMESRT, ISO 21815 apply; coordination managed internally	ISO 17757, EMESRT, ISO 21815 apply; ISO 23725 coordinates across vendors	ISO 17757, EMESRT, ISO 21815 apply; human oversight provides additional safety layer

What This Means for a Mine Making Decisions in 2026

Consider a mid‑tier operator with a mixed haul fleet across two or three OEM brands.

Integrated ecosystem: Standardize (sooner or later) to maximize capabilities; gain simplicity and strong support, but recognize that switching costs accumulate across data, processes, and people as adoption deepens.

Open platform: Deploy one AHS across the existing fleet; accept higher upfront integration effort and clearer accountability plans, but retain flexibility on vendors and data. Standards like ISO 23725 provide the technical contract that keeps FMS and AHS aligned across vendors.

Middle ground: Start with task‑bounded autonomy (e.g., autonomous drilling with remote supervision), build organizational competence, and phase decisions. This defers the architecture choice, which may resurface later under tighter constraints.

These are not abstract trade‑offs. They show up in your next RFP, in data clauses of service contracts, and in five‑year capital plans — whether or not you call them "architectural decisions."

Before You Issue the RFP (Operator's Checklist)

• Data governance: Who owns raw and processed operational data? Specify export formats, frequency, retention, and no‑penalty data egress.

• Integration accountability: If multi‑vendor, who is the named systems integrator for safety case, validation, and performance SLAs?

• Interoperability requirement: Do you require conformance to published interoperability standards (e.g., ISO 23725 for FMS–AHS API) or an equivalent documented interface contract?

• Retrofit path: Can autonomy be staged via retrofits to existing trucks or drills to de‑risk rollout?

• Human‑in‑the‑loop operations: Define supervision roles, competency standards, and control interfaces (e.g., CAP for drilling).

• Change management: Training, role evolution, and union/HR considerations for autonomy support roles.

• Upgrade cadence: How are feature updates validated across all layers (machine, autonomy, FMS, AI assistant)?

• Exit ramps: What are the switching costs and timelines if you change autonomy providers or FMS in 3–5 years?

Ground Truth from Q1 2026

1. Caterpillar & NVIDIA @ CES: Cat AI Assistant; edge inference with Jetson Thor; Riva speech in the loop.

2. HCM/LANDCROS @ ConExpo: AHS demonstrated at F19012; open/mixed‑fleet positioning; LANDCROS direction previewed (brand transition effective April 2027).

3. Epiroc live autonomy link: Las Vegas show floor ↔ autonomous SmartROC D65 at Luck Stone (Virginia) under CAP supervision.

4. Komatsu Smart Quarry Autonomous: Retrofit‑capable, staged deployment for quarry trucks; finalist for ConExpo Next Level Awards.

5. Caterpillar's aggregates milestone: 1 million tons hauled autonomously at Luck Stone's Bull Run by July 2025 (a first in aggregates), supporting the case for quarry‑scale autonomy.

What ConExpo 2026 Told Us

ConExpo didn't declare a winner. It clarified the field. For the first time, operators can see three distinct architectural answers laid out with real demos, real deployments, and real trade‑offs. That clarity is valuable: it lets mines choose deliberately instead of inheriting an architecture by default through equipment procurement. The industry isn't converging; it's differentiating. The terms of that differentiation are getting sharper. That's the development OA will be tracking: not which machines drew crowds, but which philosophies were on display and what they imply for how operations will run a decade from now.

The taxonomy exists. The shared language for evaluating it is still being written. Consider this a contribution to that language — a starting point, not the final word. These models are lenses for making trade‑offs visible, not boxes that constrain how any single operation must work.

Which architectural model best describes your operation today? And is that by choice, or by default?