I certainly struggled with the title of this article. It all began as a discussion around Telsa’s “RoboTaxis” and “Solving FSD.” Both of these concepts I find to be vague to the point of ridiculousness, but they are becoming so commonly used, AND IT IS DRIVING ME CRAZY! To have a foundation for that discussion, I decided we first need to discuss the SAE Levels of Autonomous Driving.
It seems that the SAE 5 6 levels of autonomous driving have been the go-to description of autonomous driving, but the levels are pretty odd if you ask me. Below is a quick summary, but go to the source to read more.
- Level 0: No Driving Automation
- Level 1: Driver Assistance
- Level 2: Partial Driving Automation
- Level 3: Conditional Driving Automation
- Level 4: High Driving Automation
- Level 5: Full Driving Automation
Those levels almost seem ok, but once they try to lay in examples, you see how these levels will be difficult to match to the real world. I’m not going to go too deep into this but will just point out a couple of absurdities:
The difference between level 1 and level 2 is that level 2 has both steering and accelerator/brake control. Yet, individually steering OR accelerator/brake control would be level 1. Why does having both at the same time take you to another level? Odd if you ask me.
Level 3 seems to be everything in level 4, PLUS the option to allow a human to take over. Wait, that does not make sense!
A More Reasonable Description of Autonomous Driving
First of all, the MOST IMPORTANT feature of autonomous driving is, “Who is responsible for the vehicle’s driving?”
There are essentially 2 parties involved: the human driver (if any), and the “computer” (if any). By “computer,” I think this will come to mean the company allowing and certifying the autonomous driving features used. But, this is a new huge legal item that will need to be pinned down.
In SAE’s levels, it is pretty clear that the human is responsible in levels 0, 1, and 2. And the “computer” is responsible in levels 3, 4, and 5. I find it pretty funny that they present their level 3 with a stipulation that the “computer” can pass control to the human driver. This is just unusable because it seems to indicate that the “computer” has the ability at any time to pass the driving responsibility to the human. If that is the case, then has the “computer” really ever been granted control?
Categories of Automation
Let us take a better approach to the problem of categorizing automation. Here are my easy-to-use levels categories of automation. They are based solely on who is responsible for the vehicle’s actions.
- No Automation: This is the old fashion car without any computer controls.
- Computer Assisted: The human is responsible.
- Computer Managed: The “computer” is responsible.
In each of these categories, you can have a checklist of features that are or are not available. So, to try to map back to SAE Levels.
- Level 0 is No Automation.
- Level 1 is Computer Assisted with either brake/acceleration OR steering, but not both.
- Level 2 is Computer Assisted with both brake AND steering features.
- Level 3 is Computer Managed with the ability to fail and drop to Computer Assisted so the computer is no longer responsible. (Crazy)
- Level 4 is Computer Managed with constraints.
- Level 5 is Computer Managed without constraints. (Essentially impossible in the foreseeable technical future.)
Check: Do these categories make sense to what is going on in the real world?
Current State of Autonomous Vehicles
Today, there are basically 2 approaches to adding self-driving or autonomous driving features to cars. There is the Waymo, Cruise, etc approach which is clearly Computer Managed with severe constraints. When the system fails, the company supplying the cars+automation needs to come out and fix things.
And there are the Tesla and Comma.AI approaches which are clearly Computer Assisted. For example, from Tesla’s Website: “The currently enabled Autopilot, Enhanced Autopilot and Full Self-Driving features require active driver supervision and do not make the vehicle autonomous.” Let’s dig in just a bit deeper.
With Tesla, they sell a feature labeled “Full Self Driving” that clearly lays out a set of features that it includes. The most advanced feature is the “Autosteer on city streets” which is currently in beta testing but has been released to over 100,000 drivers who purchased this feature and requested access. Nowhere does Tesla state that the car will become responsible for the vehicle, so ALL these features clearly fall into the Computer Assisted category. (This may leave you wondering about the Tesla RoboTaxis, I will get to that in the next posting.)
If you look at Comma.AI, you will see that it is somewhat similar in what it is trying to achieve but just less ambitious in terms of ability. But the Comma.AI system is available as an aftermarket “development kit” that can be used on vehicles from ALL the major manufacturers’. See: https://comma.ai/vehicles.
If you look at Waymo and Cruise, you see that the vehicle is in full control of the driving experience, so these are easily in the Computer Managed category. They just have very tight constraints on where they will work. This includes geographic regions and only specific roads within those regions. They can also be limited in speed of operation, time of day, and not available based on weather or other.
A vehicle in the Computer Assisted category would need to add assistance features to make it more like an autonomous vehicle. But, no matter what, the driver is ultimately responsible for the vehicle.
A vehicle in the Computer Managed category, would strive to remove the constraints of the vehicle’s operational environment to make it more useful. For example, allow it to drive on more roads or with fewer restrictions. Removing these constraints make it more useful as an autonomous vehicle.
Next Step: Vehicles that can move between categories
The question of crossing from one autonomous category to the other is very interesting. So, far there are NO companies that I know of that have actively delivered this feature for broad use. I see this as the logical next step, at least it is the logical next step for Computer Assisted cars.
For the current Computer Managed cars (Waymo, Cruise), it is highly unlikely that they will work to allow the transition to Computer Assisted because that is not their goal. They will work to remove the barriers and allow their vehicles to drive on more roads and conditions.
Also, the transition from Computer Managed to Computer Assisted would require that you have a driver that is licensed, capable, and willing to take the responsibility for driving the car. Since this is not a requirement from the start, hard to enforce it once a drive has started.
What about crossing from Computer Assisted to Computer Managed?
First of all, there is really only one company that may be looking into doing this and that is Tesla. Comma.AI’s CEO George Hotz has stated that the current version of their tools is designed to augment the human’s driving experience and not replace it (you can get a feel for this from Lex Fridman’s Podcast #132). So let’s focus on Telsa’s autonomous vehicle program.
Tesla’s CEO Elon Musk has numerous times stated that they will soon, “Solve FSD.” I find this to be totally absurd and meaningless, but that is another story. And, he has stated that the current fleet of millions of cars that Tesla has produced and already running on the roads, will be capable of becoming these autonomous vehicles. So, this kinda implies that these cars will be able to transition from Computer Assisted to Computer Managed. Very interesting!
One does not need to think about this very deeply to see that passing the responsibility for the vehicle from the driver to the computer and back will need to be handled very delicately. Without extremely clear delineation, the lawyers will have free reign for lawsuits all over the place. So to me, this will become the next interesting advancement in autonomous driving.
Before I go on though, I must say that I think it is highly possible that Tesla could stay away from this vehicle transition from Computer Assisted to Computer Managed. They could say that the current FSD (Full Self Driving) feature (which as of this writing costs $15,000 USD), was never meant to be a Computer Managed feature. Sure users will get very advanced features like “Autosteer on city streets” as promised, but the driver is always ultimately responsible for the vehicle. And Tesla would go on to produce a specific Computer Managed RoboTaxi vehicle that is never allowed to be controlled by a non-Tesla employee. It would be similar to the Waymo and Cruise vehicles.
Another possible path though is that Tesla could allow its vehicles to transition from their current Computer Assisted condition into Computer Managed with a caveat that it is allowed to pass back the responsibility in some very well-specified manner. This would be very interesting and seems to be within the realm of possibility.
For example, once you have engaged “computer management” of the car:
- You are free to remove your attention from driving the vehicle.
- You must be able to take back control of the vehicle within X seconds of the vehicle requesting you do so.
- If you do not take back control of the vehicle after being requested to do so, the vehicle will attempt to move to a safe location and after that, you will be GIVEN the responsibility for the vehicle.
- If you further do not take back control of the vehicle within Y seconds, an emergency call to 911 will be made to local authorities giving your vehicle location and description and any applicable data (like failure mode).
The Ridiculous World We Live In
I find it ridiculous that I need to write this article. I thought about writing it years ago, but since that is nowhere near my job, I figured someone else would work this out. But here we are in late 2022 and no one has done it.
Perhaps someone with true clout will read this, say “Oh Yeah,” steal my ideas and get the credit for it. Such is life. I hope this helps us move more quickly to a sustainable future. Good time to be alive.
Eric