How Kodiak Trained Its Driverless Tech To Haul Triple Trailers

Artificial intelligence is often defined by what it can do on a screen – generate text, create images and offer assessments. These advances have been incredible.

But AI’s next era transcends pixels. Physical AI shapes how objects move in the real world and navigate in a safe, efficient and reliable way, often over long distances. 

For Kodiak, Physical AI is already a commercial reality, and we continue to pioneer new advances while moving freight today.

We recently launched driverless technology that hauls triple trailers in collaboration with our customer, Atlas Energy Solutions. “Triples” are Kodiak’s latest industry first, and they’re unlike any other autonomous vehicle on the road today. 

They represent one of Physical AI’s most novel and compelling breakthroughs. From the front bumper to the rearmost trailer, the entirety of the vehicle stretches 50 meters, more than half the length of a football field. The loads are enormous: when full, the whole vehicle configuration can weigh 275,000 pounds, more than the weight of a fully loaded Boeing 757.

These triple-trailer trucks are now plying routes as part of a fleet of 35 driverless trucks with no humans in the cab as of June 30, 2026, and they’re owned and operated by Atlas in the oil fields of the Permian Basin. 

The Permian is an area of West Texas and Eastern New Mexico that comprises the heart of America’s oil and gas industry. Hauling triples in this rugged and desert-like environment represents a major technical advance for our team and a logistical unlock for Atlas. 

Atlas uses driverless trucks to move vast quantities of frac sand from its depots to well sites. Atlas has a relentless focus on completing deliveries on a firm schedule and maximizing the efficiency of its logistics network for the benefit of its customers. 

Without driverless technology, “triples” are difficult. Holding a standard commercial driver’s license alone is insufficient for this work. Triple-trailer drivers must meet requirements which include a special class of license and an additional ‘T’ endorsement they gain by passing additional tests.

These drivers are in short supply, especially in a remote location like the Permian where it is already difficult to find and retain drivers.

Driverless triple trailers allow Atlas to substantially increase its asset utilization – they haul 50 percent more sand than double trailers, a major efficiency gain.

Achieving this feat required training AI to calculate and manage the compounding physical uncertainties involved in safely and consistently transporting triples.

Atlas needed triples, so Kodiak delivered

The Kodiak Driver first began hauling double trailers only last fall. When we turned our attention to triples, we quickly realized it represented a far greater challenge. 

Here’s why: 

‘Triples’ are a linked chain of three trailers and two connector dollies that exhibit their own distinct articulations and behaviors. In reality, maneuvering triple requires accounting for how a tractor hauls these five independent entities.

The challenge is the vehicular version of the tail potentially wagging the dog  – and understanding why the rear trailer moves the way it does is not straightforward. 

The trailers’ behavior is further influenced by road features like the cattle guards and massive potholes common on Permian roads. Or as it is said down there, “potholes the size of a Cadillac.”

At the outset of this effort, we identified six areas we needed to consider before hauling triples: perception, planning, controls, simulation, systems and offboard infrastructure. The effort spanned the entire engineering team, and brought solutions that have contributed beyond triples and the Permian, improving the Kodiak Driver in our highway and defense verticals, as well.

Disciplined engineering gets the job done

Perception: Fifty meters is a long distance to track the position of a third trailer with the fidelity and resolution required for pinpoint accuracy, particularly in the complex path of wide turns and narrow oil and gas roads.

We upgraded our perception system, then verified it worked both in  simulation and with realistic, closed-course testing that ensured the Kodiak Driver detected other road actors who veered too close to the rear trailer. 

Planning: In the Permian, trucks regularly pass through gates and over cattle guards, narrow metal bars that bisect dirt roads.

The trick is guiding a 50-meter vehicle through a six-meter gap. The planning system must account for the path of all three trailers simultaneously and position the vehicle so that every segment passes cleanly through the opening.

“Finesse” isn’t the first trait you think of when operating triple-trailered trucks in the harsh and unforgiving Permian environment, but it is exactly what is required.

Controls: Starting and stopping this configuration requires choreography. 

Brake application cascades backward from the tractor through the dollies and trailers. It takes about one-and-a-half seconds for brake pressure to reach the rear, and our AI-powered autonomous driving system must compensate for this delay. Braking pressure and timing must be precise. Anything less is a sure-fire way to burn through brake pads.

Likewise, putting 275,000 pounds into motion without wearing out the clutch demands foresight and new dimensions of control, particularly when applying the throttle on a hill.

Systems: The Kodiak Driver doesn’t only need to account for length while driving. It must also consider mass. When full, the complete triple-trailer vehicle weighs as much as 275,000 pounds. When they’re empty, it is considerably less, a mere 75,000 pounds.

Our autonomous driving system must understand the weight it carries on any given job. Braking harder than a lighter load needs introduces instability throughout a triple-trailer configuration. Brake less than a heavier load requires, and the trucks might overshoot their desired stopping point.

Load specialists input the number of trailers into our operations software before every run, and Kodiak has created a mass estimator that determines the weight of each load on each axle. We then track how well our actual braking profiles compare with what the system expected, and our operations team is notified if that performance falls outside a specified performance threshold.

We’ve also added a “trailer count estimator” that independently verifies the number of trailers on the actual truck matches what’s expected. In case of an input error, this backstop catches mistakes and ensures resilience.

Offboard infrastructure: Not every route is appropriate for triples. Routes are constrained by a number of factors, including the radius needed to cleanly make all turns from its departure point to destination. 

Kodiak built an offboard system that checks for restrictions before a trip ever begins. Every route run with a triple-trailer configuration must be explicitly approved for that activity. The system tracks approval status, expiration dates and ensures they only operate in authorized environments.

Safety first, through Probabilistic Risk Assessment

Performance is secondary. Safety is first. 

All our work to bring triple-trailers to the Permian has been vetted through our Probabilistic Risk Assessment, a safety methodology Kodiak uses to estimate collision risk, quantify edge-case uncertainties and prove the safety of our driverless technology in statistical terms.

The PRA represents an iterative, ongoing process that updates as we accrue new data and experience, which we’re gaining every day in our 24/7 active work with Atlas.

Since launching triples in February, the Kodiak Driver has carried more than 1,400 loads in the triples configuration.

The Western Star 49X, and the road ahead

In August, we took another step forward. Atlas recently selected the Western Star 49X as its newest driverless platform. The Kodiak Driver is now hauling triples using the 49X, seamlessly integrating our AI-powered software and modular hardware with a heavy-duty machine built for the toughest conditions.

Atlas is committed to increasing the size of its Kodiak-powered driverless fleet to 100 trucks by the middle of next year. 

Hauling triples represents Physical AI pushed to its limits. Building this capability felt like a master class in mass, momentum, kinetic reaction and unpredictable environments all mixed together – using AI to solve problems too vexing for most humans.

This breakthrough also reflects the dedication and discipline of the Kodiak team. We take pride in tackling the hardest problems. When Atlas first asked if we could solve this daunting logistics puzzle, we said, “Not yet.” But we accepted the challenge. We divided the problem into manageable steps. We executed and delivered. 

And the tools we built – across perception, planning, controls, systems and offboard infrastructure – are contributing toward our target of launching driverless trucks on public highways by the end of this year. 

That is the real promise of Physical AI. It is bringing real, measurable value to the entire logistics industry that powers the world and its economy.

David Silver leads planning and controls at Kodiak AI and spearheaded development of “triples.” He has worked on autonomous-driving technology for the past decade, and can be reached on LinkedIn.