Defining a Safe Driver
As I’m getting ready to fly to the ATS 2025 conference in Tempe, Arizona, I still see new posts about AV industry experts asking the now age-old proverbial question: “How Safe is Safe Enough?” The answer to the difficult question lies in how we frame the problem by asking: “How Reasonable is Reasonable Enough?”
What is Reasonable driving? This we can answer and define. We have mountains of data to show what Reasonable is. But statistics are not enough. So, we use behavioral models of Reasonableness backed by good, self-evident explanations as to why some drivers are “Reasonable” and some are “Unreasonable.” You can turn left in front of on-coming traffic. The question of how Reasonable or Unreasonable depends on when we turn left in front of on-coming traffic. These can be defined using clear, unambiguous physics-backed rules. No AI. No complicated statistical comparisons. Just common sense.
When our 7-year old and 10-year old go to our neighborhood park, my wife and I remind them to be careful of cars and to look both ways. Our children can sit their observing the motion of vehicles from a distance and tell if the car is coming towards them. They can tell if the car is quick, slow, turning left, turning right, stopping, etc. They don’t have any advanced sensors monitoring the pedals and steering wheels, they’re just watching the basic vehicle motion (we call “bicycle-model” steering) and it is clear to see within a few hundred milliseconds where the drivers are going over the next few seconds.
As a parent, I could watch my children interact with cars in our neighborhood and judge my children’s decisions of “Reasonabless” on some sort of qualitative scale, which I can quickly back up with quantified conditions: How much distance? How fast are they? How fast are you? How much time? How visible are you? How much space is there to navigate around?. The scale could be boolean: “Reasonable” or “Unreasonable” or it could be gradual, with 3, 4, 5, or even 100 degrees of “Reasonableness” based on my preferences. Typically, with Safety Management Systems for autonomous vehicles, we need several degrees to detect an indicator’s warning and respond with sufficient resources.
What we are excited to share with the industry, and especially our public safety stakeholders, is that “Safe Driving” does not need to be some complicated, proprietary metric. No AI. No complicated statistical comparisons. Just common sense. We have tools we call Control Action Analysis to determine the fundamental behavior that drivers are projecting based on uninformed, outside observations (sometimes there can be multiple potential behaviors and that’s ok - it doesn’t have to be perfect. It has to be reasonable).
But these algorithms and measurement methods do not need to be proprietary or controlled by us to add value - quite the contrary. The more they are understood, tested, and improved on by everyone in the community then the faster we can have commercial autonomous operations which pose no Unreasonable risk to the public (see what I did there?).
Besides the behavioral reasonableness models, we have great quality, naturalistic datasets to help us develop and verify our models. Our friends at leveLXData (LXD) have the highest quality, real-world datasets of real drivers and road users interacting with each other, using high precision aerial drone surveys. In fact, without their data progress on reasonable behavior models would be severely hampered. This is why most AV companies have either heard of leveLXData or use their data which has been collected from Arizona, Austin, Ann Arbor, Aachen, Germany, and all over the world.
Please book a consultation with us if you’d like to help make universal safety metrics available for all. We’ve included a small snapshot from Barton Springs and Lamar in Austin, Texas data that was collected in July 2025 to show the thousands and thousands of high resolution behaviors tracked in the real-world and analyzed according to useful behavioral reasonableness models. How Safe is Safe Enough? Reasonably Safe is Safe Enough.