Autonomous heavy equipment and construction automation

Scope and source note

This note covers the autonomy stack for heavy construction and mining equipment: perception, planning, control, safety architectures, fleet integration, and the commercial deployment models emerging around autonomous excavators, dozers, haul trucks, and compact earthmoving equipment.

I distinguish granted patents from applications, and peer-reviewed papers from preprints. A patent is evidence of recorded claims, not proof of performance or commercial viability. Deployment claims are cross-referenced against company filings where possible.

Technical architecture and performance claims

1.1 Autonomy stack

Heavy-equipment autonomy generally decomposes into:

• Perception — LiDAR, radar, stereo cameras, and GNSS/RTK localization.
• Mapping and localization — 3D terrain models, obstacle detection, and localization against a digital reference map.
• Planning and control — path planning, excavation/trajectory generation, and closed-loop control of hydraulic actuators and powertrain.
• Supervision and safety — a human supervisor either on-site or remote, with fail-safe stop logic and interlocks.
• Fleet management — mine- or job-site-level dispatch, blending, collision avoidance, and data aggregation.

This stack is visible in Caterpillar’s autonomy product architecture, which uses an onboard autonomous control system, off-board mission planning, and a command center for human oversight Caterpillar MineStar Command product page (accessed 2025-06-13). Caterpillar’s autonomy system is deployed on mining trucks through the “Command for hauling” product, which the company describes as enabling fully autonomous operation in mine environments, with an operator monitoring from a control center Caterpillar MineStar Command product page.

The technical foundation for excavation autonomy is documented in a peer-reviewed paper:

• Paper: “An autonomous excavator system for material loading tasks,” Science Robotics, 2021, DOI: 10.1126/scirobotics.abc3164 Science Robotics (published 2021; accessed 2025-06-13). Peer-reviewed, DOI-verified.

The paper reports a complete autonomous excavator system using vision and LiDAR perception, terrain-aware planning, and an integrated control architecture for material loading tasks. It is the strongest independent Tier 1 performance source in this note.

Cross-verification status for the core technical claim that autonomous loading is feasible: The Science Robotics paper and Caterpillar’s deployed mining autonomy lineages are independent institutions. However, the specific claim “autonomous excavator loading matches expert human productivity” currently depends on the Science Robotics paper and supporting field-test reports. The paper is single-source for the exact experimental setup, unless a second independent benchmark can be retrieved [SINGLE-SOURCE for exact productivity ratio]. The general claim that autonomous excavation is technically feasible is cross-verified by Caterpillar/Komatsu/other OEM field deployments, so I do not treat the entire thesis as single-source.

1.2 A granted patent record

Patent: U.S. Patent No. US 9,475,520 B2, cited here as a recorded control-system claim in the machine-automation class. The title/assignee/grant date should be independently retrieved from USPTO or Google Patents before any legal or moat conclusion is drawn; I could not fully resolve the front-page metadata in this environment. It is marked needs primary source validation for the exact assignee and grant date. URL used for the record: Google Patents US9475520B2 (accessed 2025-06-13).

Why a patent appears early in this note: Patents in this class cluster around control methods, perception fusion, and safety interlocks. The patent record is evidence that assignees are claiming rights in the technical space; it is not evidence of field-proven autonomy, reliability, or commercial uptake.

1.3 Safety standards

• ISO 17757:2017 — Earth-moving machinery and mining — Autonomous and semi-autonomous machine system safety. ISO (published 2017; accessed 2025-06-13). This is a Tier 1 standards-body document. It defines safety requirements for autonomous and semi-autonomous machine systems, including functional safety, perception degradation, human-machine interfaces, and site-level safety management.
• The standard is relevant because many deployments in mining are justified partly by removing workers from dangerous zones, but the safety argument depends on validated stop-and-failover behavior. ISO 17757 is the closest independent regulatory-adjacent benchmark for that claim.

1.4 Company filings cross-reference

Caterpillar’s SEC filings disclose risk factors around technology development and customer capital budgets without making specific autonomous-product revenue promises. The company’s most recent 10-K filings can be retrieved via EDGAR:

• Primary filing access: SEC EDGAR: Caterpillar Inc. 10-K filings (accessed 2025-06-13).
• In Caterpillar 10-K risk factors, the company discusses, among other risks, the timing and customer acceptance of new products and technologies, cybersecurity vulnerabilities, and the competitive environment in autonomous mining and construction equipment. Those filings do not provide autonomous-specific revenue splits or margins; they treat autonomy as one element of the broader equipment and services business.

Reported discrepancy / tension: Marketing pages from OEMs and autonomy providers often emphasize 24/7 availability and safety improvements. The 10-K, by contrast, emphasizes customer acceptance, capital-budget cycles, regulatory/legal uncertainty, and cybersecurity risk. There is no direct numerical contradiction in the filings reviewed, but the marketing frames should be read as operational upside claims, not filed financial commitments. No autonomous-specific revenue line could be verified in Caterpillar’s 10-K from the EDGAR index URL above; segmented disclosure is at a coarser level. This is a knowledge gap for margin attribution.

Deployment and adoption context

2.1 Mining haul trucks

Mining haulage is the most mature deployment segment. Komatsu’s Autonomous Haulage System (AHS) page states the system has been used in large surface mines and has moved cumulative tonnage milestones. This is company self-reported Tier 2 evidence. Komatsu AHS (accessed 2025-06-13).

Caterpillar’s MineStar Command page similarly describes autonomous hauling in mines, with on-board and command-center operation. Company self-reported. Caterpillar MineStar Command (accessed 2025-06-13).

Note: Both are OEM self-reported sources. Independent mine-level tonnage-cumulative figures should be sourced from a mine operator or regulator, not from the OEM alone. The Komatsu and Caterpillar claims are two independent OEM sources, so the deployment class is cross-verified at the category level; specific tonnage figures are not independently verified in this note.

2.2 Construction retrofits and supervised autonomy

Built Robotics uses a robotics-as-a-service retrofit model for excavators and skid steers. The company’s public technology page describes autonomous path planning, obstacle detection, and off-site supervision. Company self-reported Tier 2. Built Robotics (accessed 2025-06-13).

Teleo advertises supervised autonomy for heavy construction equipment, where a remote operator can take over and operate one machine or supervise multiple machines. Company self-reported Tier 2. Teleo (accessed 2025-06-13).

These sources support the existence of a retrofit autonomy market, but specific revenue, unit counts, or uptime metrics are not independently verified. Treat all specific productivity claims from these pages as marketing until cross-checked against a neutral benchmark or pilot operator.

2.3 Construction equipment operator labor data

U.S. Bureau of Labor Statistics provides a stable public source for construction equipment operator wages. It is not an autonomy claim, but it is the input for unit-economic calculations. BLS Occupational Outlook Handbook: Construction Equipment Operators (accessed 2025-06-13). The exact median annual and hourly figures should be confirmed on the BLS page; in the worked calculation below I use illustrative values and label them as assumptions.

Unit economics / business model

3.1 Who monetizes, and how

• OEMs — Caterpillar, Komatsu, Deere, CNH Industrial, Volvo: monetize by selling autonomy-capable machines, software licenses, fleet-management subscriptions, and service contracts. The autonomous upgrade can be bundled into the machine sale or sold as a technology package. Company self-reported via product pages; exact software price lists are not public in SEC filings reviewed.
• Retrofit autonomy providers — Built Robotics, Teleo, SafeAI: monetize via RaaS/hourly or per-job fees, leasing autonomy kits, or charging per supervised autonomous hour. This shifts customer capex to opex. Company self-reported Built Robotics (accessed 2025-06-13).
• Fleet operators — Rio Tinto, Fortescue, BHP, large contractors: capture value through reduced shift costs, improved utilization, lower collision/downtime, and the removal of workers from hazardous zones. Their unit economics are often presented in investor days but rarely isolated as “autonomous-only” margin in SEC filings.

3.2 Margin characteristics

No autonomous-specific margin was found in the SEC filings reviewed. Caterpillar’s 10-K reports segment-level financials, not autonomy-specific subscription margin. The closest analog is Caterpillar’s broader “Resource Industries” or “Construction Industries” segment profitability, but attribution to autonomy is not disclosed. Therefore, a typical autonomy gross margin cannot be sourced from Tier 1 filings without additional primary-source retrieval. needs primary source validation.

The software/subscription nature of fleet-management autonomy suggests higher incremental gross margin than hardware manufacturing, but this is an inference, not a sourced financial fact.

3.3 Who captures the value

• In mining: the mine operator captures much of the value through labor productivity and safety; the OEM captures hardware/software revenue and multi-year service contracts.
• In construction retrofit: the RaaS provider captures a recurring fee; the contractor captures productivity gains if autonomy reduces headcount or increases utilization.
• In both cases, the purchaser is the fleet owner/operator, not the individual machine operator. The customer decision usually requires a capital or project-economics sign-off, because the machine must still be integrated into existing site workflows.

Source-bound caution: The above business-model description is drawn from OEM/retrofit product pages and 10-K segment disclosures, not from an independent cost study.

Worked quantitative calculation

5.1 Objective

Estimate annual labor savings from replacing one worker-shift of direct operation with supervised autonomy on a single construction machine, and show sensitivity.

5.2 Formula

Annual direct labor savings = hourly loaded labor cost × annual operating hours × labor-share displaced.

Where:

• hourly loaded labor cost = base wage + benefits + overhead per worker-hour.
• annual operating hours = planned machine hours per year.
• labor-share displaced = fraction of direct operator attendance that hands over to autonomy/supervision. This is not 100% for supervised teleoperation, because a remote operator may monitor multiple machines.

5.3 Inputs and assumptions

5.4 Calculation

Annual direct labor savings = \(36.4 × 2,000 × 0.7 = \$50,960\) per machine-year.

5.5 Sensitivity

• If base wage is $20/h, loaded cost = $26/h → savings = \(26 × 2,000 × 0.7 = \$36,400\) per machine-year.
• If base wage is $40/h, loaded cost = $52/h → savings = \(52 × 2,000 × 0.7 = \$72,800\) per machine-year.
• If labor-share displaced falls to 0.4 at $36.4/h: \(36.4 × 2,000 × 0.4 = \$29,120\).
• If labor-share displaced rises to 1.0 at $36.4/h: \(36.4 × 2,000 × 1.0 = \$72,800\).

Interpretation: A retrofit autonomy kit or software subscription must cost materially less than roughly $35k–$75k per machine-year to create direct savings on a single-shift basis, unless utilization gains or safety/insurance benefits are separately quantified. This is a sensitivity, not a sourced market price. The actual kit price and subscription fees are not filed in SEC disclosures reviewed. needs primary source validation for kit/subscription price.

Parameter table

Disconfirming Evidence

The thesis that autonomous heavy equipment and construction automation is a durable, scalable value-creation sector would be wrong or materially weakened if:

1. No independent Tier 1 performance benchmark replicates the Science Robotics excavator result in a different environment or material type; the current paper is single-source for exact productivity parity.
2. OEM autonomous-specific revenue or RaaS subscription revenue fails to grow in Caterpillar, Komatsu, or retrofit-provider financial disclosures; if autonomy remains a line-item marketing feature without margin disclosure, the unit economics are unproven.
3. A major mining or construction pilot is abandoned or a regulator restricts autonomous earthmoving operation on safety or cybersecurity grounds; this would undermine the adoption catalyst and shift buyer economics.
4. The patent record is shown to be non-controlling or largely expired in key jurisdictions, reducing moat and pricing power for autonomy incumbents.
5. Labor displacement assumptions are materially lower than 0.7, or remote supervision proves to require near one-to-one operator attention, collapsing the direct savings estimate toward zero.