#1 cleantech news, reviews, & analysis site in the world. Subscribe today. The future is now.

Autonomous Vehicles

Published on January 1st, 2016 | by James Ayre


Level 4 Autonomous Driving Possible In 2 Years, Elon Musk Contends

January 1st, 2016 by  

Originally published on EV Obsession.

Level 4 autonomous driving will be possible within only 2 years, according to Tesla CEO Elon Musk — as quoted by Fortune in a recent interview. This comment comes as something of a contrast to some (but not all) of Musk’s previous statements on the matter — most of which placed fully autonomous driving off a couple of years further into the future.

Despite the apparent conviction that the technology will be possible within only 2 years, Musk did note in the recent interview that it will take (at least) another year in order to get regulators on board for approval.


“The point at which it becomes statistically clear that an autonomous car is safer, I think, regulators will be comfortable with allowing it,” as he put it in the interview.

Our sister site Gas 2 provides more:

During that year, the software will operate in “shadow mode,” comparing what it is programmed to do with what actually happens in real world driving. Using artificial intelligence algorithms, it will teach itself how to deal effectively with what Musk calls “corner cases.” Those are the millions of unknown and unknowable situations that human drivers face every day. The distracted pedestrian who dashes into traffic unexpectedly. The skateboarder who completes a perfect Ollie but lands in the middle of a crosswalk. The moment when a car runs a stop sign or slips sideways on an icy road.

Tesla has already turned its cars into one of the most advanced artificial intelligence networks in history. All of its cars with the Autopilot suite of sensors and software installed learn from their daily driving experiences and share that knowledge with each other. If there is road construction in Terre Haute, your Tesla will know about it, even if you live in Walla Walla and are driving in Indiana for the first time. That’s exactly the kind of learning Musk expects Tesla’s Level 4 systems will do while working in “shadow mode,” waiting for regulators to turn them loose on public roads. He expects the network to provide the data regulators will rely on when deciding Level 4 autonomy is safe.

That’s about the approach that I’ve been expecting the company to end up taking, so I can’t say that I’m surprised by Musk’s comments. It certainly seems to be a sensible approach to getting the regulators on board. Hopefully it works!

Tags: , , , , , ,

About the Author

James Ayre's background is predominantly in geopolitics and history, but he has an obsessive interest in pretty much everything. After an early life spent in the Imperial Free City of Dortmund, James followed the river Ruhr to Cofbuokheim, where he attended the University of Astnide. And where he also briefly considered entering the coal mining business. He currently writes for a living, on a broad variety of subjects, ranging from science, to politics, to military history, to renewable energy. You can follow his work on Google+.

  • phineasjw

    Well Google already has 1 – 2 million miles on their fleet, which is level 4 autonomous driving.

    Speaking of which — keep your eye out for an announcement in three days (January 5th). There’s a strong rumor that Google will announce a partnership to build autonomous cars with FORD at the Consumer Electronics Show (CES).

    Fully autonomous driving is coming. The more interesting question, I think, is how soon will human driving be outlawed?

  • Autonomous vehicles have been around for at least 10 years. In 2004 DARPA Grand Challenge offered $1,000,000 million dollars prize money for the first robot to make a 130-mile course through Nevada’s desert terrain. That attempt failed miserably. In 2005 they offered it again but doubling the prize money (2 million). What I found very interesting is the tale about the robot called “Stanley” built by Stanford University. In testing, they set up a course for Stanley to run. The course had two rows of Jersey barriers. In between those barriers were road cones. Stanley’s job was to drive a slalom route between the cones as fast as he can. Now this is what gets interesting. You would expect the robot to do the run as predicted. This is where Stanley outsmarts human drivers. Stanley scans the course then backs up and positions himself between the Jersey barrier and the cones. The speed though the course in record time. You see Stanley figured out he could fit between the cones and the barrier and didn’t need to do a slalom route. The engineers jaws dropped to the ground when they seen this. Yes, Stanley was programmed to find solutions, but this solution just caught them all off guard.

    Having a robot to do your driving is like having a professional racecar driver at the wheel. He’ll be a much better driver than most of us. Autonomous cars don’t have to be perfect, they just have to be a bit better than us.

    The Great Robot Race.

    Here is a video of the program: https://youtu.be/vCRrXQRvC_I

    Since 2006 there was one other race called the DARPA Urban Challenge. That race was in an abandoned Naval base. The robots had to navigate through the city with human drivers in the mix. There was not V2V communications between the robots. They had to work independently and without any human intervention as they did many routes and maneuvers.

    We are now at 2016 and these robots can drive better than you.

    • Jenny Sommer

      Only point I disagree…autonomous vehicles have to be perfect not only a little better. At least better than 95% of human drivers.

  • By early 2017, the US Department of Transportation hopes to publish a rule mandating vehicle-to-vehicle (V2V) communication by an as-yet unspecified deadline. GM says that by the 2017 model year, the Cadillac CTS will be V2V equipped.

  • Matt

    “Level 4 autonomous driving will be possible within only 2 years”
    This means that in 2 years the tech will be possible. Not that it will be used on any public roads. They will then gather data, and refine, and start the very long process of get regulators somewhere to allow it.

  • Ivor O’Connor

    That’s about the approach that I’ve been expecting the company to end up taking, so I can’t say that I’m surprised by Musk’s comments

    Is it April 1st already? These comments sound about the same as his comments about colonies on the moon in 10 to 15 years back in 2011. They sound nice but…

    Half the vehicles on the road will be 100% EV before this happens. And that is not going to happen in two to three years!

    • Carl Raymond S

      I’m going to be bullish on this one; I expect it will happen sooner rather than later.

      It’s a software project which can be divided, with no new hardware invention required. So the development cycle is super fast – write code – test on simulator – pass > commit – fail > revoke. For faster progress, add more developers. It’s counterintuitively less complex than designing/making second row seats.

      The divisibility of it reminds me of the human genome project, planned for 15 years and completed (I think) in around 7.

      I don’t see a need for cars to be able to handle every situation – so long as there is an internet connection. If there’s a mad gunman or drunkard on the road (I’ve experienced the latter), then a remote professional driver wearing a VR headset can take control. A small team of such drivers could act as standby for an entire national fleet – and still more or less eliminate driver labour cost.

      • Bob_Wallace

        Tesla has thousands of cars in the field right now collecting the data needed for the simulator runs. Every time a driver grabs the wheel and overrides the autopilot system or every time the autopilot hands control back to the driver is an event that can be added to the database. A year or so of data and one would be hard pressed to argue for the existence of events that aren’t covered.

        Airplane coming in for a emergency landing on the freeway? Rule 1) Don’t hit anything. Rule 2) Stop if that keeps the car out of trouble. Rule 3) Drive away from a potential impact if there’s someplace to drive.

        What more could a human driver do? Would their decision making process be as rapid and objective as the computer’s?

        Sniper in the woods? I’m not sure either computer or driver would make the right call. At least with autonomous driving the driver could dive to the floor and let the car continue on out of the danger zone.

        • Carl Raymond S

          Tesla have already had the joy of discarding the engine, allowing all those savings to offset the battery/motor. It’s deja vu all over again. For the taxi model – toss out the steering wheel, the driver console, the mirrors and the pedals. Marginal cost of an autonomous car will be less than the cost of an existing car.

        • I agree. Autonomous cars don’t have to be right 100% of the time, just right more often than humans.

          • dRanger

            What’s going to be interesting is when the shadow mode shows all the times that the computer would have taken a course that would have avoided an accident.

  • nakedChimp

    would had been nice to give a wrap-up of what level X is supposed to mean.. not all follow these kind of developments that closely.

    • Andy

      Fully atonomous, as in take a nap in the back seat. Not that regulations would ever allow for it. Look how long it’s taking them to allow cameras instead of side mirrors.

      • Carl Raymond S

        Some ‘maps’ stop at level 4. They should add level 5, where the cars talk to each other and drive like one of those precision driving teams that perform at shows. Crazy fast and way too close. Sleeping will be a good low stress option.

          • Carl Raymond S

            Well, I might have to edit that wiki and add level 5. :-]

          • Clearly you didn’t read it. 😉

            “Level 4: The vehicle performs all safety-critical functions for the entire trip, with the driver not expected to control the vehicle at any time. As this vehicle would control all functions from start to stop, including all parking functions, it could include unoccupied cars.

            An alternative classification system based on five different levels (ranging from driver assistance to fully automated systems) has been published by SAE, an automotive standardisation body”


            Basically both Level 4 and SAE Level 5 are the same.

          • Carl Raymond S

            I understand what level 4 is – the car replaces the driver. I am saying there is a level above that, where the cars communicate with one another, so that each car knows in advance what what other cars are going to do. Sensors can only reveal what other cars are doing right now. At this communicative level above level 4 (which I am calling level 5), cars can pass very close, traffic lights become redundant and traffic can move fast, even at high densities.
            We won’t see this higher level until some time after all cars on a given road road reach level 4. Perhaps 15 years from now.

          • V2V isn’t necessary. Humans have been driving cars for years without radioing in their intentions. Surprisingly aircraft pilots and ship captains all radio their intentions. That said, V2V will help robots make quicker decisions and even be able to speed up traffic by coordinating actions. V2V isn’t really a level change because the vehicle is already fully autonomous.

            I predict that there will be a time that no one will be allowed to drive manually unless it’s on a closed circuit like a racetrack. How long will it take us to get there? Well it will take 20 years just to replace the entire fleet of cars we already have. Even Elon Musk mentioned that. What will hurry the adoption is that people will find cost savings in that lower or no insurance. The ability to go out drinking and not having to drive home. Having your car drop you and your family off at the front door of a football game while it goes off and finds parking by itself.

          • Carl Raymond S

            OK, you have a name for it – V2V, and you say that V2V will speed up traffic by coordinating actions. So call it level 4 plus V2V, or call it level 5 – I’m not going to quibble over a name.

            Imagine 4 streams of traffic approaching a roundabout. With V2V, they can position themselves perfectly on approach, then zipper their way around the roundabout, then continue, slowing only enough to avoid an uncomfortable G force.

            Remove V2V, and you have all cars stopping and politely waiting for somebody else to go first. It’s chalk and cheese, and this is just one of hundreds of scenarios. Level 4 is not the end of vehicle automation.

Back to Top ↑