The SAE levels of autonomy are broken, it’s time for a rethink as we approach level 5

Update
6 days after writing this article, the SAE levels of autonomous systems was indeed updated.

SAE International is an association that focuses on developing standards for engineering professionals across various industries. Once of their most high-profile pieces of work is the Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles J3016.

Originally created back in 2014, it was revised in 2016 and again back in 2018. The standards were set out to describe how we think about autonomy, with each car offering a different suite of technologies, this enabled some level of comparison between manufacturers. The problem is, as we get closer to some automakers moving beyond levels 2 and 3 on to 4 and 5, the current levels are not aligning with how the autonomous story is progressing.

In theory an automaker would progress through each one of the SAE levels, but in reality, you could actually develop technology to jump from what we think of as Level 2 (partial automation), to Level 5 (full autonomy). More commonly what we see is that cars ship with level 2 and receive software updates to improve capabilities that place them somewhere between level 2 and 3, let’s say 2.5.

While we can all agree level 5 refers to the ultimate goal, where humans have been completely replaced and we could get into vehicle that never requires intervention, or attention from us. This opens the door to many new, exciting new possibilities, like going to sleep, being drunk, using our phones, watching movies, doing work or turning around and interacting with our kids in the back seat.

So let’s take a look at the levels as they were presented back in 2016, a chart many people still circulate today.

What are the SAE Levels of Driving Automation?

Take any car and try to position it in the column for level, 2, 3 or 4, good luck. The issue is that some functionality is at level 2 (say city streets), some functionality is at level 3 (say highway driving). This is complicated by other vehicles like autonomous shuttles (like the one from Navya) don’t have wheels and pedals, but also don’t drive on the freeway (and often not on public roads).

As an effort to address this, SAE updated the table to what we see below which aims to make it easier to understand when a driver is required and when they’re not, but this simplification, does nothing to address the fundamental issue.

As we lean into a point where regulatory approval will be required to enable autonomous cars on our roads, having standards should enable those regulators to test the capabilities of a proposed tech stack and determine if its safe.

With vehicle’s capabilities varying based on low, medium and high-speed driving, it doesn’t seem like SAE levels really cut it in today’s environment. The next challenge is the software one, where a car could be classified as safe for use with certain conditions, but with a new version of software, increase its capabilties, but unless our regulation is prepared to be dynamic, would artificially restrict what an automaker could offer with a software update.

This is an international challenge and I’d love to hear your thoughts in the comments about how we can navigate this minefield.