The Birth of the AV — the Autonomous Vehicle
This article is part 1 of 4 in a series written by Turing Motor Company’s CEO: Ned Goodhue. This series outlines the history and Turing Motors' approach to Artificial Intelligence (AI) and its role in autonomous vehicles.
Despite billions of dollars and over a decade of work, we have yet to see an AV team that is stronger than level 3, or even level 4 — but in a geofenced area with limitations, never mind level 5+ fully autonomous vehicles.
In fact, as of early 2020, over $16 billion has been spent on AVs. (1)
For example, Google’s AV team, Waymo, started in 2009 and despite raising billions, has failed to deliver fully autonomous vehicles and mass-scale adoption.
Some of the remaining active teams are:
- Z00X — It started in 2014 and was bought by Amazon in 2020. Their goal is to create vehicles to be a robo-taxi. It is bi-directional. Amazon also envisions the robo taxis being used to deliver packages.
- Cruise — they started as a division of GM, but now they have other investors. If you look at https://www.getcruise.com/ you can see what they are doing.
- Waymo — started as a Google project (Chauffeur), like Cruise have taken outside investors recently. https://waymo.com/.
- Tesla — It is difficult to know where they really are: beyond Advanced Lane Detection? For years they have said they would be delivering fully autonomous self-driving by the end of the year, however, one of their most recent updates took away radars and several features from some of their models.
There are a few teams who are creating the technology that would be put on a car to make them “AV”.
Argo AI is a great example. https://www.argo.ai/
Udacity has one, too. https://www.youtube.com/watch?v=gFLx6D6zmUY
The Race to Trucks — for revenue and to make a profit on their investments.
Many of the AV teams discussed above have failed to raise many revenues, much less profit.
Surviving teams have turned their focus and money to trucks. Why?
- Fewer variables. Transport trucks largely only drive on highways, predictable routes, and face less unexpected variables than cars.
- To have a big truck to go from one warehouse to another warehouse is far easier than training autonomous vehicles to drive in all conditions and all possibilities
- More profitable for logistics companies. Logistics companies will not have to pay for a truck driver or worry about regulations that limit the number of hours a driver can drive without breaks. The drives would also be more consistent and predictable for coordination
It is a smart decision. Embark, Aurora, Daimler, Einride, TuSimple, Volvo, Tesla, and Waymo are all looking to be the leader in the truck AV market.
But they have failed for autonomous cars and public transportation
How Come?
They used the technology that had. Lidar. Radar. All sorts of sensors. And put them to work. Silicon Valley loves it.
https://arberobotics.com/ Look at the start of this company’s URL.
This is a great radar company in Israel. It shows what all of the teams have been doing for the last 11 years.
And look what has happened?
Failure.
It is easy to see.
Let us start again with a completely different approach:
AI — Artificial Intelligence
There are many forms of AI. But each of them tries to:
- Understand how people do something or get to some goal, then AI…
- Uses specific approaches to get a better or more reliable answer using a different approach than a person would use.
- The result, we hope, will be a faster and more accurate answer while keeping the same goal as a person would want.
Example: Playing Chess.
There is a company in London, UK, called DeepMind. They started in 2010. Google bought them in 2014. They do pure research. https://www.deepmind.com/.
Using a variant of AlphaGo Zero (their Software & Hardware), DeepMind:
- Taught the rules of chess to two parallel copies of AlphaGo Zero. (This is the technology part of it. It knew nothing but the rules).
- 8 hours later, AlphaGo Zero was playing chess at Grand Master level.
The final report was published in “Science” in 2018. (https://www.sciencemag.org/news/2018/12/google-s-deepmind-aces-protein-folding)
The key takeaway was that the Software and Computers did not simply turn on some software that played chess at Grand Master level. It was the software that got smarter and smarter, not the programmers.
It is a very different concept to understand — some software and fast computers who, on their own, learned to play against humans, and win. In some way, a human wrote some software that teaches itself to be the best players in the world, in a few hours.
That is Artificial Intelligence, and those who learn to do it have the tools to go right to Level 5 in an AV.
Example: Predict how a car behind us will behave.
Coming down to the world of Autonomous Vehicles, let us show you a piece of our (Turing Motor’s) technology — pure AI — that a driving software can predict the action of that car behind us.
A person: Experienced drivers on a freeway can and will look behind them and notice a car coming up to in our left-hand lane. The driver will create a strategy almost instantly.
No software can do that — as we have seen so dramatically on the streets of San Francisco and other towns. It is the reason that every AV team has failed.
But Artificial Intelligence can. Here is a short description of how our driving model works.
Turing’s AV will be able to drive better than the best professional driver.
Why? That is impossible.
Because a human can know more and faster than any computer. But they can have only a few objects in the brain. It is simply how our brains are built.
But Autonomous Vehicles can keep 25, 50, etc. objects in its “brain” at any given time. It is never distracted. It won’t be talking with a friend for example. So the process of AI may be very different, but the result is better.
And it starts at Level 5 as it should.
“Any technology, sufficiently advanced, is indistinguishable from magic”
Arthur C. Clarke
We are the first of the next generation — success with AI.
This article was written by Ned Goodhue. Ned is the Founder and CEO of Turing Motor Company. Ned is passionate about all things AI. Ned is also passionate about aviation, travelling, understanding different cultures, theatre, and reading.
Feel free to leave a comment, whether you agree or disagree, we encourage a collaborative discourse. Thank you.
-Turing Motor Company
Index
Algorithm — A well-defined steps that a computer can do that the author wants. It can be a result or an action.
Analytics — a tool or solution
Attribute — a quality or characteristic of a person, place, or thing
Bombe — the machine that Alan Turing made at Bletchley Park to break the code of Germany’s Enigma code. The nickname of Bombe was “Victory.”
Calculable — It is possible to come up with an accurate answer in time to be useful.
Incalculable — The answer will be either wrong or it arrives too late to be useful.
Predictive — foretelling the future. Especially useful in insurance, in a war, or predicting what that car over there will do. This is sometimes called Predictive Analytics.
Proactive — A car can be smart enough to tell us something is going wrong before it fails. That goes with parts, too.
Reactive — waiting until something fails.
Restore — to bring (one of our used cars) back to perfect condition. We will do it with all our cars every 3–4 years.
References
- Baldwin, R. (2020, February 10). Self-Driving-Car Research Has Cost $16 Billion. What Do We Have to Show for It? Car and Driver; Car and Driver. https://www.caranddriver.com/news/a30857661/autonomous-car-self-driving-research-expensive/
