This is Post 1 of our four-part series exploring how SHM™ can help the ADS industry comply with the recently released UN GTR for Automated Driving Systems. You can find the intro post here.

Ask anyone what a competent and careful driver looks like, and you'll get a fairly consistent answer, probably not much different from the one provided by the UK Government: someone who follows the rules, pays attention, adjusts to conditions, and looks out for the people around them. Nobody argues about the concept. The recently published UN GTR on Automated Driving Systems leans its entire safety bar on exactly that phrase: an ADS must perform at least as well as "a competent and careful human driver”. But they don’t really specify any further, not because regulators disagree on what "careful" means, but rather because turning it into something you can actually measure has proven harder than expected.

Or rather, it looks like no one has managed to measure it successfully. UNECE, the United Nations Economic Commission for Europe, published guidelines in May 2025 that, along with a multipillar testing approach and the need for a safety case, offered two driver models (or Mathematical Kinematic Models) that attempt to quantify the issue: The Competent and Careful Driver Model (CCDM) and The Fuzzy Safety Model (FSM). Both formulas define kinematic thresholds (using Time Headway and Time-to-Collision) that describe what a careful driver should do at one specific moment, like the instant a neighbouring car starts drifting into your lane. That's a reasonable starting point, as we have argued that kinematics is the right raw material. But a formula built around one moment is a hypothesis about careful driving, not a measurement of it. It has to guess the right thresholds and delays in advance, then hope real driving agrees with the guess. Eventually, when both CCDM and FSM were empirically tested (Olleja et al., 2025)¹, the results showed that “if models are to be included in regulations, they need to be substantially improved.”
‍

We believe we need a more direct route. Rule adherence and attentiveness aren't things you test for once—they show up continuously, as hazard exposure stays low across thousands of ordinary interactions, not just the ones that almost end badly. Adaptability shows up as that same low exposure as conditions worsen (heavier traffic, poor visibility, wet roads) instead of a fixed threshold quietly becoming the wrong (dangerous) one. Consideration for pedestrians and cyclists shows up directly in how closely and how fast a vehicle passes them, interaction by interaction, not as a rule bolted on afterwards.
‍

That's the shift SHM™ makes: instead of specifying what a careful driver should do at a moment, it measures what one actually does, continuously, across every moment. Point it at a population you'd genuinely call competent and careful (a fleet with a clean record, licensed driving instructors, drivers in a specific ODD) and the resulting distribution of hazard exposure isn't a hypothesis. It's the definition, built from the population it describes, rebuildable for any ODD, any fleet, any baseline you need.
‍

"Competent and careful" doesn't need a better guess. It needs a population to measure, and a tool like SHM™ to turn it into a meaningful number.

‍

\\\

Original photo by Selcuk Sarikoz on Unsplash

‍
¹ _{Ollejaa,P., Markkula, G., Bargman, J.  |  Validation of human benchmark models for automated driving system approval: How competent and careful are they really?  |  Accident Analysis and Prevention 213 (2025) 107922  |  PDF Link}

‍

‍