Safetronic 2026: Preview
From certified port autonomy to AI-enhanced defense vehicles

The journey began with a bounded operational design domain (ODD): container transport in ports. Controlled access, trained personnel and established operating rules made the risks manageable, but the environment was far from simple. Heavy vehicles shared space with workers, drivers interpreted lane markings pragmatically, and trailers swept through tight turns, while rain, fog and reflecting puddles challenged perception. These interactions shaped safety functions whose required integrity did not exceed Performance Level d.

To make the safety argument tractable, Fernride separated Machine Learning (ML)-based nominal operation from an independent, deterministic safety layer. Its hazard zone monitor covered the full volume the tractor and trailer could occupy during an emergency stop, including the safety-enforced motion range, uncertainties and system latencies. Together with speed and steering-rate limits and optical and acoustic pre-start signals, this provided a safety baseline on which operational capability could evolve. Making this work reliably outdoors nevertheless pushed deterministic technology to its limits.

Complex Certification Process

Certification rested on an integrated safety and security lifecycle, combining ISO 13849 for system and hardware safety, IEC 61508-3 for software integrity and the IEC 62443 series for cybersecurity. Coupled hazard and threat analyses, supplier assessments and a Git monorepo tied the machine configuration to its evidence. Approximately 800 hazards and 370 compliance documents supported a safety-critical implementation of around 5,000 lines of code. Automated software, resimulation and Hardware in the loop (HIL) tests, vehicle validation, and final-design Electromagnetic Compatibility (EMC) and mechanical testing completed the assurance effort.

That achievement raises a new question: what happens when autonomy leaves the terminal? Defense applications range from rear-area road logistics to smaller unmanned vehicles crossing difficult terrain, where mud, water and vegetation demand scene understanding and road traffic requires predictions of other participants’ motion. ML, including end-to-end architectures, offers a route to these capabilities, but changes the evidence needed before deployment.

Residual risks must be assessed

Here, ISO/PAS 8800 and SOTIF help structure the next stage of the safety argument. Systematic ODD coverage and empirical scenario coverage guide data curation, while independent validation datasets and scenario testing in validated simulation and on the vehicle provide evidence for incremental capability extensions. Yet the long tail of rare scenarios leaves residual risks and uncertainty to assess.

Defense operations offer additional ways to address that challenge. Authorities can separate traffic, restrict access and train personnel intensively, reducing exposure to difficult scenarios, while mission conditions influence which residual risks may reasonably be accepted. Protecting soldiers and civilians remains essential, because avoidable accidents undermine morale and operational capability and because skilled personnel are difficult to replace. Manufacturers must make residual risks and uncertainties transparent enough for an informed mission-specific decision.