It took some ten years for the heavy construction industry to fully adopt GPS/ GNSS earthmoving technology. Now medium-to-large contractors can鈥檛 work profitably without it.

A similar trajectory is likely with the recent introduction of artificial intelligence (AI). The changes that are coming promise big leaps in productivity and efficiency, but a lot of work has to happen first.

AI encompasses a broader trend called machine learning, and the hard work is going to be teaching these machines everything they need to know. As it turns out, even the smartest machines are no match for seasoned operators and site supervisors 鈥� at least for now.

John Deere revealed their autonomous articulated dump trucks (ADT) at the recent Consumer Electronic Show 2025 in Las Vegas, US. (Image: John Deere)

Start small, scale up

鈥淭here seems to be a shift of focus from when we were going all-in on autonomy to taking a series of steps to get more automation that one day will lead to autonomy,鈥� says Ian Welch, Trimble鈥檚 engineering director with civil construction field systems.

鈥淗uman intelligence is not just data. It is our intuition,鈥� says Burcin Kaplanoglu, head of Oracle Industry Labs. 鈥淚t鈥檚 hard to beat someone who has spent two decades in construction. But that is the promise. AI is coming and we are embedding those things as features into our products.鈥�

The goal itself is simple and not unlike the goal of GPS/GNSS guided earthmoving; to make less-skilled operators better and skilled operators faster and to do the job safer, better and greener.

AI will play a role in figuring out to optimise a process, says Welch. 鈥淭hat鈥檚 step one in the journey,鈥� he says. Step two is more difficult. 鈥淲hat do you do when things don鈥檛 go as planned?鈥� AI will need to make real-time decisions, and I think we鈥檙e still a way off from that.鈥�

Phased approach

Neil Williams, president Machine Control Division, Hexagon鈥檚 Geosystems division, says the construction industry has transitioned from the initial hype around fully autonomous construction equipment to a more practical, phased approach. This includes five levels of autonomy:

1) Manual operation 鈥� Full human control with minimal digital assistance.

2) Assisted control 鈥� Automation aids operators with guidance and automatic blade control.

3) Partial automation 鈥� Machines handle some tasks autonomously but need operator oversight.

4) Conditional autonomy 鈥� Machines operate mainly independently in specific conditions with minimal human intervention.

5) Full autonomy 鈥� Equipment functions entirely on its own, adapting in real-time (not yet an industry standard).

Currently, the industry is progressing through levels two to four, while keeping operators in control, says Williams. Instead of full machine autonomy, the industry is moving toward more semi-autonomous workflows, where machines collaborate more seamlessly with operators and leverage AI to enhance productivity and reduce errors.

Neil Williams, president Machine Control Division, Hexagon鈥檚 Geosystems division

This makes automation a foundational tool that transforms operations, setting the stage for full autonomy in the future while delivering tangible benefits now, says Williams. These include enhanced machine control, AI-driven analytics; and realtime data integration to streamline workflows, boost safety, and cut waste.

According to Hexagon鈥檚 2023 Autonomous Construction Tech Outlook 84% of technology decision-makers at general contracting firms in North America, the UK, and Australia had adopted some form of autonomous technology in the previous year, says Williams.

Autonomy isn鈥檛 a distant goal, comments David Veasy, John Deere鈥檚 senior product manager of autonomy. But it will come in focused autonomous operations. These autonomous operations will be simple at first, such as hauling material from point A to point B, and grow in scale over time, focusing on minimising the amount of interaction needed to keep the machine in the autonomous mode.

John Deere revealed their autonomous articulated dump trucks (ADT) at the recent Consumer Electronic Show 2025 in Las Vegas, US. ADTs in a quarry application carry material from point A to point B and repeat that route without too much variation, making it one of the ideal applications for an autonomous machine. Given a shortage of skilled laborers, the autonomous ADT eliminates the operator and enables customers to use that person for more complex tasks and equipment.

Simple machines first

The machines most likely to incorporate autonomy in the future will be the ones with the simplest applications, says Welch. Several years ago, Trimble developed the software and sensors for a semi-autonomous compactor the company frequently demonstrates.

Burcin Kaplanoglu, head of Oracle Industry Labs

Compactors were the logical choice because they only do one thing, and it is easy to proscribe that path and optimise that process, he says.

An excavator is more complex and is also called on to do many different tasks in different conditions, says Welch. Likewise with dozers.

鈥淲e are a way off from automating those machines now but there are a lot of people working on it. We鈥檒l see more of it happen as we see digital construction become more widespread. But without that we can鈥檛 really train the machine.鈥�

鈥淵ou can get to 90% autonomy, but that last 10% is hard to achieve,鈥� says Kaplanoglu. And we, as humans, have a very low tolerance for machine failure and fault. We are tolerant towards human mistakes, but not machine mistakes. Consistency, reliability and accuracy is pretty important in the construction space.鈥�

To see and then do

At the Oracle Industry Lab, the team uses OCI Vision Services to analyse photos and images and then trains the software to recognise elements in the images and compile the information needed to utilise that asset.

A hypothetical example he cites is when an HVAC (heating, ventilation and air conditioning) unit shows up on a jobsite without documentation. A quick photo sent to the cloud can identify the unit, and if properly trained, the software will not only identify the unit, but conjure up the specs, installation instructions and any other relevant information. Enhanced by AI learning the software will produce all the information required, integrated into one response without the worker having to go to multiple websites or open multiple screens of information to get what they need.

Some companies are taking that visual learning model a step further and programing robots to 鈥榮ee鈥� a particular action, learn it, and then repeat it.

The tech term for this is 鈥榠mitation.鈥� Kaplanoglu cites an example where a robot records professional athletes and then recreates those movements itself.

The implication is that it may be possible in the future for a dozer to record another dozer performing a task, store that information in the cloud and then use that information to duplicate the task without an operator in the seat. 鈥淭hat, I think, is the biggest promise,鈥� says Kaplanoglu. 鈥淭hat is where it鈥檚 heading and there are tons of money and research going into this space.鈥�

Upskilling your crew

鈥淎utomation is skilling jobs, not killing them,鈥� says Williams. By taking over repetitive tasks, technologies allow workers to focus on higher-value tasks, from decision-making to process optimisation. This not only improves efficiency but also makes jobs more engaging and helps bridge the skills gap. Companies that embrace automation aren鈥檛 cutting workers 鈥� they鈥檙e making them more valuable, he asserts.

As machines and jobsites evolve to incorporate more autonomy, the operator鈥檚 job may transition to becoming a technology manager, says Welch. Skilled operators are still going to be needed to teach the machines how to perform certain tasks. 鈥淎nd you will get to a point where there are really hard, really complex tasks, with lots of inputs that only a human can manage. So, it will be many years before we can do everything without a human in the cab,鈥� he says.

Swarming micro machines

David Veasy, John Deere鈥檚 senior product manager of autonomy

It is also possible, maybe even likely, that autonomy could change the design of earthmoving machines other than eliminating the cab, or changing the number or size of machines to do a particular job, says Welch.

鈥淭here is some interesting research looking at whether it would be more efficient to have two large machines or ten small robotic machines that work together.鈥�

Trimble calls this 鈥榤ulti-machine coordination鈥� says Welch. Instead of two big dozers it might be more like ten mini dozers all working as a swarm at the same time. OEMs will drive this development, but technology, software and machine learning will play a big part.

Autonomy will also enable a single operator to control more than one machine, says Deere鈥檚 Veasy. The trend over time will be that autonomous equipment will require less and less operator engagement to operate and manage.

The impact of AI

Automation existed long before AI was a household name. However, AI is now the hottest ticket in town and is and building on the success of GPS/GNSS machine control.

AI has unlocked the ability to create productive autonomous solutions that previously would have taken much longer to develop and would have had several limitations to their usefulness, says Veasy. For example, using computer vision to decipher between objects and types of objects is only achievable in a commercial way because of the advancement in AI.

Autonomy will likely change the look and design of equipment (Image: AdobeStock)

Williams says that AI hasn鈥檛 changed Hexagon鈥檚 direction but has enhanced what the company were already doing. Machine learning, digital twins and automation have been improving precision, efficiency, and decision-making for years. What鈥檚 evolving now is AI鈥檚 ability to scale, process data faster, and provide more actionable insights in real-time.

While the future of AI and automation are cutting edge, there鈥檚 no reason to take a wait-and-see approach. Incremental improvements with real world benefits are happening almost every day. 鈥淲e鈥檙e integrating AI to refine machine control, automate data processing in surveying and reality capture, and enhance site monitoring and progress tracking 鈥� not to replace operators, but to help them work more efficiently with greater confidence,鈥� says Williams.

In the final analysis, the rate of adoption will vary by the technology and benefit. 鈥淲e have seen customers adopt technology at a high rate, when the value provided outweighs the costs,鈥� says Veasy. The historical example he cites is GPS/GNSS grade control.

鈥淥nce customers experience that they can get to grade in one pass, the time savings and efficiency outweighs the initial cost investment,鈥� Veasy continues. 鈥淲e expect that as the cost for advance tech is reduced and the need for more productive equipment increases, technology adoption will scale accordingly.鈥�