Driverless trucking moves from demo miles to production contracts

A 500-truck deal and a factory line

Aurora Innovation announced its third quarter 2026 results with an emphasis on high-volume production and commercial scaling. Charger Logistics signed a non-binding memorandum of understanding to own and operate 500 Aurora Driver-powered trucks, with deliveries expected in 2027 and 2028. The filing describes the arrangement as a Driver as a Service model, and says the memorandum represents a potential multi-year revenue stream in the hundreds of millions of dollars. The memorandum is non-binding, so it is a statement of intent rather than a firm order, but it is the largest demand signal in the announcement. m:…4168 m:…4710

Aurora has more than 20 driverless trucks already built, has surpassed 500,000 commercial driverless miles since launch, and expects to reach 20 trucks per week in early December. It is integrating the Aurora Driver directly on the Volvo factory line, and it is working with Roush to establish an annualized production capability of more than 1,000 trucks. Aurora ended the third quarter with more than 1 billion dollars in cash and short-term investments, which funds the production ramp while the Charger deliveries are still a year or more away. m:…4168 m:…4710

The filing is candid about a constraint: the initial production ramp over the past month was constrained by material availability and component quality issues. That sentence matters more than the headline numbers, because a production story stands or falls on whether factory output can be sustained week after week. The Federal Motor Carrier Safety Administration also granted a five-year exemption permitting driverless trucks to use cab-mounted warning beacons instead of manual roadside devices, which removes one regulatory friction from driverless operation. m:…4710 m:…4168

The data loop behind the trucks

Applied Intuition describes the data flywheel that turns fleet driving into better models. Its flywheel processes hundreds of petabytes of multimodal data and grows by more than a petabyte a day. Curation selects segments for training and evaluation datasets based on metrics derived from sensor and vehicle signals, and video-text embedding models are used to find driving scenarios not yet described by metrics or labels. Training scenarios consist of windows of up to 300 aligned frames, stored in Lance tables that allow scenarios to evolve without rewriting files. m:…1727

Training runs in two stages: imitation learning followed by reinforcement learning. Imitation learning teaches the model to drive like an expert by matching human driver paths, and reinforcement learning allows the model to learn from the consequences of its own actions in simulation. Evaluation suites include both open-loop replay of scenarios and closed-loop driving in simulation. The sequence from curation to training to simulated evaluation is the industrial part of autonomy: the model improves because the loop runs continuously, not because a single drive went well. m:…1727

A safety case, checked independently

May Mobility has completed more than half a million commercial autonomous rides, and Edge Case has completed the first phase of an independent safety assessment of May Mobility safety case framework. Edge Case applied its DevSafeOps methodology, evaluated the framework for structure and completeness, and evaluated alignment with UL 4600 and the Open Safety Case standard. Edge Case found the framework consistent with industry best practices, in an assessment of a partnership that began in 2022. m:…7210

May Mobility is now streamlining its safety case framework for easier interpretation and management, strengthening its safety performance measurement against a human-driving benchmark, expanding its monitoring and evaluation of real-world driving data, and formalizing its release decisions to include independent assessment. Its physical AI system integrates deep learning, a dynamic world model and a real-time reasoning engine. The assessment is only a first phase, but the direction is clear: safety claims are being documented, benchmarked and checked by an outside party, rather than asserted by the operator alone. m:…7210

Why production changes the question

This suggests the question for autonomy has changed. It is no longer whether a truck or a shuttle can drive a route without a driver. It is whether a company can build trucks at a weekly rate, sustain that rate through material and quality constraints, sign multi-year operating deals, run a data loop that improves the model every day, and present a safety case that an independent assessor has checked. Aurora, Applied Intuition and May Mobility each answer a different part of that question, and the companies that answer all of them together will be the ones that scale. m:…4168 m:…4710