Key To Driverless Cars, Operational Design Domains (ODD), Here’s What They Are, Woes Too

Dr. Lance B. Eliot, AI Insider

Image for post
Image for post
Need standardized definitions for Operational Design Domains (ODD) for self-driving cars

When I got my very first car, I was so excited to be driving my own car that I opted to drive everywhere that I could think of.

I drove all throughout my local neighborhood and honked my horn as I drove past the homes of friends of mine. I drove beyond my community and took the freeway to go visit friends that lived in the inner-city areas. I drove down to the beach, parked at the edge of the sand, and took a picture of me and my car, including as a backdrop a dramatic sunset and the rays of the sun glinting off the ocean, adding a picturesque look to my shiny new automobile.

The next day, I took some friends up to the snowy mountains. I had made sure to buy snow-chains for my tires and it was my first foray into putting them on and seeing how the car handled on icy roads and in light snowy conditions.

Upon returning from the mountains, a good friend suggested we head out to the desert. The mountain trip had involved freezing cold temperatures. Perhaps by going to the dry and hot desert, we’d be able to unfreeze and gain back our normal body temp. With my still brand-new car, we drove on a somewhat barren highway and headed out to the middle of the desert.

Not being content with “only” having driven in the city, the suburbs, the mountains, and the desert, I decided that the next adventure with my car would be to the forest. So, I packed my camping gear into the car and drove up to the redwoods.

As I headed back home, the forest got deluged with quite a rainstorm. Driving out of the woods was a bit tricky as some of the roads began to flood.

One other thing that I encountered was a lot of potholes and other marred roadway aspects. Because of the rain on the road, I was not able to readily discern where the cracks and gaps in the asphalt existed.

Each of these road trips involved encountering varying kinds of roadway conditions. There are some places that I could drive, and some places that I could not readily drive, often times dictated by the nature of the road or the weather conditions or other factors. And, limited too my own driving ability (I’m somewhat squeamish about driving on icy roads, which we don’t get much in Southern California).

Let’s refer to the various roadway conditions as a type of domain, one in which the car and driving is considered operational, and by-design the car — and me, the driver, are able to drive presumably under those conditions.

You could say that those are Operational Design Domains, usually abbreviated as ODD.

Operational Design Domains Defined, Kind Of

In the official parlance of the automotive industry, the way in which you can define the scope and limits of driving are referred to as a “domain” and commonly indicated as the Operational Design Domain (ODD). Per the IEEE standard known as J3016, here’s what ODD formally means: “… operating conditions under which a given driving automation system or feature thereof is specifically designed to function, including, but not limited to, environmental, geographical, and time-of-day restrictions, and/or the requisite presence or absence of certain traffic or roadway characteristics.”

That’s a bit of a mouthful.

In essence, an ODD is a kind of carve out. Imagine all of the numerous ways in which driving might occur such as in fine weather or bad weather, on bumpy roads or smooth roads, and so on. In that universe of a myriad of driving conditions, you can stake out a subset and declare it to be an ODD.

You Can Make Up Your Own ODDs

I might define an ODD that consists of smooth roads, absence of rain and the roads must be dry, and there must be high visibility in terms of being able to see around the car. That’s my declared ODD. It’s just one such ODD. I might define a second ODD, for which it consists of smooth and bumpy roads, light rain allowed, roads can be wet but not slick, and the visibility can range from high to mediocre.

I could continue declaring various ODDs. Each of the ODDs would have some particular set of indicators about what it includes. This might also include exclusions, thus I can probably be clearer about my ODDs by not only saying what it includes but also what it excludes. That being said, the number of exclusions could be rather vast and perhaps exhausting to try and list them all.

There is no accepted standard as to what the ODDs are.

Think of the number and variety of ODDs that could be declared. By mixing and matching all permutations and combinations of the myriad of factors, you could create an enormous number of ODDs. Besides the roadway aspects, you can state that an ODD is good for daylight but does not encompass night time. Thus, time of day can be a factor. The geographical area can be a factor, such as I might declare my first ODD was intended only for say Los Angeles and no other geographical realm.

On and on this can go.

Lack of Standards for ODD and AI Self-Driving Cars

What does this have to do with AI self-driving cars?

At the Cybernetic AI Self-Driving Car Institute, we are developing AI software for self-driving cars. For the Level 4 and Level 5 of AI self-driving cars, the nature and use of ODDs is essential. Once the public begins to experience Level 4 and Level 5 AI self-driving cars on the roadways, the ODDs topic is going to hit the big time and be at the forefront of public discussion and discord. Mark my words!

I’d like to first clarify and introduce the notion that there are varying levels of AI self-driving cars. The topmost level is considered Level 5. A Level 5 self-driving car is one that is being driven by the AI and there is no human driver involved.

For self-driving cars less than a Level 5, there must be a human driver present in the car. The human driver is currently considered the responsible party for the acts of the car. The AI and the human driver are co-sharing the driving task.

Another key aspect of AI self-driving cars is that they will be driving on our roadways in the midst of human driven cars too.

Returning to the topic of ODDs, let’s take a closer look at what they are and why they are going to be so crucial to the advent of AI self-driving cars.

Level 4 and ODDs Are Essential

Level 4 self-driving cars must provide an indication of the ODDs under which they are able to operate.

This means that if you are intending to purchase a Level 4 self-driving car, you would be wise to look carefully at the ODDs that the auto maker says are applicable to the automobile you are about to purchase.

The reason to scrutinize the ODDs is so that you’ll then know where, when, and under what circumstances your AI self-driving car is going to be able to operate as a self-driving car.

According the standard definition for Level 4, once the AI detects that it has reached a point that the driving is no longer within its defined ODDs, the AI is supposed to let the human driver in the car take over or the AI is supposed to pull over, finding hopefully a safe spot to do so, and wait to continue driving until the situation becomes one encompassed by the ODDs of that particular AI self-driving car.

Suppose you going along for a spin in your fancy new Level 4 AI self-driving car. It is a wintery day. When you began your joyful journey, the skies were relatively barren of clouds. Sure, it was a cold morning as you got underway, but you didn’t expect bad weather to occur. Darned if toward lunch time, clouds started coming in fast. With the cold temperatures and the clouds forming, it begins to snow.

The AI of the Level 4 self-driving car is presumably able to detect the snowfall, doing so via the sensors of the self-driving car. Because the ODDs indicated the AI is not considered able to drive in snowy conditions, the AI alerts you that you’ll need to take over the driving of the self-driving car. If you refuse or don’t speak-up, the default will be to pull the self-driving car over to the side of the road at the earliest feasible and hopefully safe spot.

Even though you might be able to drive the self-driving car, and you are willing to do in spite of the flakes of snow, and there’s lots of other car traffic around you that is doing so, your AI is not going to budge one inch. The ODD boundaries have been reached. You would need to take the controls if you didn’t want to sit there by the side of the road and wait for whenever next the snow cleared up sufficiently that the AI declared it was okay for it to proceed and would continue on the driving journey.

I realize you might say that it is a small inconvenience in this case. No big deal, you say, it’s a minor annoyance that the AI has opted to no longer drive the self-driving car for the moment. For your driving journey, at least it drove you for a substantial part of the time. You can just now take over the driving and finish the trip. Furthermore, if you are able to drive out of the snowy area, you likely can coax the AI to resume driving the self-driving car.

But, imagine that you decided to have your Level 4 AI self-driving car take the kids to school that morning. You had put the kids into the self-driving car and sternly instructed the AI to drive them straight to the school. It had done this many times before.

Unfortunately, on this particular day, let’s assume that the snow starts to fall from the sky while midway to the school. The AI announces that it needs either a licensed driver to take over the controls right away or it will pull over to the side of the road. There isn’t a licensed driver in the self-driving car (you are still at home, awaiting the self-driving car to drop the kids at their school and come back to pick you up to drive you to work). Only your underage children are in the self-driving car. They can’t legally drive and nor do they know how to drive.

Regrettably, they now are going to be sitting in the dormant and roadway-parked AI self-driving car which has found hopefully a safe place to sit out the snow. For most parents, this would be a chilling moment and the time at which they start to have second-thoughts about having gotten that Level 4 AI self-driving car.

My overall point is that the ODD’s of the Level 4 AI self-driving car could consist of a wide variety of inclusions and exclusions. I made things over-simplified by using just the snowy condition. There might instead by a large number of inclusions and exclusions, making it much harder to judge when you might have the AI opt to quit on you. It won’t be so easy that you’ll readily be able to predict when the exclusions are going to be reached.

Ridesharing Impacts Due To ODDs

The other way in which the ODDs will impact you is when you use a ridesharing service.

It is predicted that ridesharing services will flock in droves to using AI self-driving cars. This makes a lot of sense for the ridesharing firms to do so.

You are getting off work early and decide to take a ridesharing car to get home. Via a mobile app on your smartphone, you summon a ridesharing car, doing so with a company that prides itself on providing all and only AI self-driving cars. A few minutes later, the AI self-driving car comes to the curb and you get into it.

Unbeknownst to you, this particular Level 4 AI self-driving car has an ODD that the auto maker and tech firm defined to exclude heavy gusts of winds. Once on the highway, you can see up ahead that the winds are shoving trucks and other cars

All of a sudden, the AI announces that you will need to take over the driving or it will be pulling over to the side of the road momentarily. Why, you ask? The AI responds that the wind speeds are excessive and exceed the defined ODD for this AI self-driving car. Yikes! You had no idea that this ridesharing car had that kind of an ODD. You are irked to no end.

Each Auto Maker Can Define Their Own ODD

Making matters murkier, each auto maker can define ODDs in whatever way they might prefer.

This means that you won’t have any apparent or immediate way to determine for any given AI self-driving car make or model what kind of ODD’s it assumes.

It is going to be the wild, wild west of ODDs. A potential boondoggle.

There will likely be an ODDs war, perhaps auto maker X says their ODDs are better than auto maker Y. This will go on for quite some time. The usual leap frogging of high-tech and automobiles will of course occur. One year, auto maker X will have a “better” ODD set than auto maker Y. Meanwhile, with continued innovation and advancement, the next year the ODD set of auto maker Y might be better than the ODDs of auto maker X. This will continue, over and over.

Some Emerging Research Is Helping to Figure Out ODDs

It is handy that some researchers are trying to help us out of this mess, doing so before the mess becomes fully evident.

For example, Dr. Krzysztof Czarnecki at the University of Waterloo has been putting together a helpful ontology for ODDs. He organizes the proposed ontology into five key areas, consisting of road structure, road users, animals, other obstacles, and environmental conditions. He defines an Operational Road Environmental Model (OREM), consisting of relevant assumptions about the road environment, and then crafts Operational World Models (OWMs) that consist of OREMs with one or more subject vehicle models. This is the kind of rigor we need to get established for ODDs.

If we could get ODDs into a more structured and agreed form and format, it would certainly make the publication and comparison of them a lot easier and more readily understood.

I’ll toss another idea out there and see what you think of it.

Suppose the auto makers and tech firms were able to compartmentalize their AI systems to correspond to the elements of a defined and standardized ODD. Besides allowing for comparing auto maker X to auto maker Y in terms of their Level 4 AI self-driving cars, we might be able to do something else too.

It might be possible to have auto maker X and auto maker Y offer to do a deal whereby there might elements of each of their respective ODDs that they could exchange with each other. Suppose that one has dealt with handling high winds, while the other one has focused on dealing with snowy conditions. They might opt to share with each other, in which case auto maker X now gets the snowy condition added to its ODD and the auto maker Y gets the high winds condition added to its ODD.

I say this with a grain of salt. A really big grain of salt.

The odds are that the hardware and the software of the AI systems of the auto maker X and auto maker Y are so vastly different that it is unlikely they could just offer up their respective components of handling high winds and of snowy conditions to each other. Instead, they each would have something extremely proprietary that works only on their own setup of hardware and AI software.


Will we be able to standardize and then possibly amalgamate ODD’s of Level 4 AI self-driving cars? The jury is still out on this.

Admittedly, it is a stretch right now to think that it could happen. There is too much momentum of each auto maker or tech firm doing their own proprietary ODDs and there is little or no incentive to do otherwise. As I’ve mentioned, if things get out-of-hand, it could be that regulators step into the morass and offer some kind of sticks and carrots to get ODDs to become more manageable.

I’ve not said much herein about Level 5 AI self-driving cars.

In theory, a Level 5 AI self-driving car is a complete set of all of the possible ODDs that would exist in a more scattered manner for Level 4 AI self-driving cars. In other words, the Level 5 is not supposed to have any curtailing limits, other than there is no off-roading capability required and also that the driving task must be something that a human driver could have managed (if a human could not have driven in the circumstance, the Level 5 definition says that there should not be an expectation that the AI could).

Some believe that to get to a Level 5, you should first make Level 4 AI self-driving cars. You could then presumably tie together all of the various ODDs over time that you crafted for the Level 4’s and make your way to a Level 5. There are those that eschew such an approach and say that you should forget about doing any of the Level 4’s. Don’t get mired into the itsy-bitsy ODD’s. Instead, aim for the whole enchilada. Aim for the Level 5.

If indeed the Level 4’s are quickly replaced by Level 5’s, it would likely lessen the impact of the ODDs wars and fractionalization. People would barely have had time to complain about the confusion over the myriad of ODDs, and presumably be quieted and more satisfied once they rode in a Level 5. For those that believe we are going to have Level 5’s on the heels of Level 4’s, the ODD topic is a “don’t care” to them. Those ODD’s will be like yesterday’s fad that came and went. Forget about it, they would assert.

I’ve said and written many times that a true Level 5 AI self-driving car is like a moonshot. I hope you can now see why I say this. The Level 5 has to be able to handle all of the permutations and combinations of all of the ODD’s that you can think of. That’s a lot to deal with. I doubt that the time gap between Level 4’s and Level 5’s is going to be as swift as some pundits claim.

I applaud those that are forging a means toward an ontology of ODDs.

We need to do more to get the ODDs matter on-track before it becomes a train that goes off the tracks. For those that are standing on the railroad tracks right now, it takes a lot of vision to see what’s going to happen miles and miles away, into the future.

Imagine if the railroads had not agreed to a common means of laying track, and upon reaching Promontory Summit in Utah on May 10, 1869, they would have not been able to drive that final golden spike into tying together the Central Pacific and the Union Pacific Railroads. That’s the kind of progress that can occur when you get your standards act together.

Let’s do the same for ODDs.

For free podcast of this story, visit:

The podcasts are also available on Spotify, iTunes, iHeartRadio, etc.

More info about AI self-driving cars, see:

To follow Lance Eliot on Twitter: @LanceEliot

For my blog, see:

Copyright 2019 Dr. Lance Eliot

Written by

Dr. Lance B. Eliot is a renowned global expert on AI, Stanford Fellow at Stanford University, was a professor at USC, headed an AI Lab, top exec at a major VC.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store