Why doesn't Tesla use LIDAR?

edited November -1 in General
Why isn't Tesla using Lidar in it's newer vehicles. Saw on KCET a discussion by a Ford Scientist that their vehicle will be using Lidar capable of seeing up to 200 meters.

They hope to have the vehicle ready by 2021.

Will cameras and sensors be enough to do all that is required of a true autonumous vehicle ?


  • edited November -1
    The standard answer is that LIDAR doesn't work in fog, whereas RADAR does. Maybe we need both.
  • edited November -1
    To me it all depends on the quality of the image data from the cameras and the success of the learning computer to understand that data.

    How do humans drive without LIDAR? We have good image data and a very good learning computer.
  • edited November -1
    Tesla doesn't use LIDAR because Elon doesn't believe it provides any benefit over the current hardware. If he ever changes his mind, we may see LIDAR in a future generation of AP hardware.
  • edited January 2017
    LIDAR has also been exceptionally expensive - in the tens-of-thousands until recently. The prices are coming down, but it still costs thousands to add Lidar to a car today. The hope is the price will come down in 2-5 years.

    Then there is the looks of Lidar - weird rotating beasts on the top of the car which look rather ugly. Again, there are solid state Lidars with limited range (i.e. you need 4+ scattered around the car to get 360 degree coverage), but its less clear if these have enough resolution. The cost is "projected" to be under $1000 per car at some time in the future.

    LIDAR also doesn't detect color or light - so you still need cameras to detect red tail lights, red/green stop lights, signs, etc. Google has proven LIDAR works (along with cameras), but it may be an unnecessary and costly approach. Radar and cameras may do a better job, as the radar work in conditions where LIDAR and cameras will not (fog, snow, heavy rain, etc.).

    Other than costs, perhaps another way to look at LIDAR is a viable approach that some may adopt, while others will use alternative systems like cameras and radars that Tesla uses. Just like some people pick a GSM phone network and others pick CDMA - both work fine, and you really choice your phone vendor on other aspects.

    Currently Tesla has more AP miles driven than any other car maker, perhaps more than all other car makers combined - so they do know a little bit about the technologies, costs and practicalities.
  • edited November -1
    Elon does not believe FSD can work without LIDAR which is expensive and intrusive to aesthetics and aerodynamics. Everything is fluid of course you never say never.
  • edited November -1
    can't work without LIDAR
  • edited November -1
    LIDAR is susceptible to malicious disruption unless you plan on spending millions and millions of dollar developing and maintaining proprietary security on your systems. Thanks to the high amount of data involved in LIDAR adding that security and prevention becomes a massive resource drag.

    This would leave the vehicle falling back on radar and cameras any time there is a possible distortion essentially nulling any benefit gained from the costs invested.
  • edited January 2017
    Tesla's sensor suite already is far better than a human's. All systems, no matter the sensors used, require a good brain to make sense of the data. Elon has read the future right. Invest in the brain.
  • edited January 2017
    If I were Google, and I were building a dozen cars and employing people to drive them around all day long to collect real-world miles, I'd put every single sensor known to man on the cars. While you're developing your autonomous driving algorithms later, you can always choose to ignore some sensors to determine the minimum set you need to work well. But if your test cars don't have, for example, LIDAR, you can never tell whether it's needed or not, or whether it would be able to resolve issues that you couldn't without it, because you don't have the LIDAR data.
    So, just because the current Google cars have LIDAR, doesn't mean that their eventual solution will need it. Perhaps they'll figure out that radar and cameras is sufficient.

    This is one area that Tesla's approach misses out. They're putting the AP2 hardware on ALL their cars in order to collect data, but all those billions of miles of data don't have LIDAR info; for all those niche cases where their cameras and radar fail, they'll have to make educated guesses as to whether or not LIDAR would have helped.
  • edited November -1

    I agree that Tesla might be missing LIDAR data if it is added to the hardware suit later. On the other hand it probably will not be that big a deal since even under that situation RADAR and camera will still be able to run the system while LIDAR is learning.
  • edited November -1

    That's a good point. Us human drivers only got at most two eyes.
  • edited January 2017
    Radar and cameras might be better once you integrate the information.
  • edited November -1

    There is a big difference between putting AP2 hardware on all cars and using it to generate data - and adding unnecessary/redundant technology to all cars just to get data in case that data is valuable.

    There is very little down-side to putting this existing hardware in all cars. In fact it simplifies manufacturing because they don't have to invest in customizing the hardware for each car. These days the cost is not in the plastic, silicon, etc. that goes into (for example) the AP2 hardware. The cost is in the development and upkeep of that technology. The development cost is the same if you put it on one car as it is to put it in all cars. There is no upkeep cost because they don't warrant to people who didn't pay for the AP2 hardware and are not using it - that it will continue to function. All they have to do is disable auto-pilot use for those that didn't pay for it. That is very easy to do.

    Whether or not Lidar has any benefits is not something you determine using an entire fleet of cars. You at least have to have a theory as to its benefits over radar. It's not good enough to say "well, Google tried it, it must be good".
  • edited January 2017
    LIDAR has much better resolution than RADAR. However, unless you have enough computing power to process the large set of LIDAR data in real time, there isn't much gain. Throwing away some of the unimportant data to analyze the data quicker is a common strategy, but there is always the danger that the data actually was important.

    Regarding RADAR and snow or heavy rain, it doesn't scatter as badly as LIDAR, but it is greatly attenuated, decreasing its range. Even if the autopilot uses RADAR, weather still matters.
  • edited November -1
    Tesla does use LIDAR on some of their engineering test cars. I've seen the twirling LIDAR cylinder on the Tesla test car roof. That's probably why Tesla can say LIDAR isn't needed. Tesla has tested where LIDAR is needed and engineered a way for the AP2 system to get adequate results.

    This is the classic use of software to replace hardware. Software is cheaper to mass produce (make copies) and update. This AP2 software must be a huge challenge.

    Since Elon and Larry Page of Google are good friends, they have probably shared a few insights on how to do this.
  • edited January 2017
    The Quanergy solid state radar is going into full production this year, not in 2-3 years time.
    It will be priced around $US250, not thousands.
    http://www.businesswire DOT com/news/home/20170103005387/en/Quanergy-S3-Solid-State-LiDAR-Worlds-Affordable
  • edited January 2017
    Solid state LIDAR is steerable and can change it's resolution, focus and frame rate dynamically. It has very high resolution. It does not see through fog and rain like radar though. I would not be surprised if Tesla adds LIDAR.
  • edited November -1
    Solid state LIDAR needs multiple sensors on a car to give 360 degree coverage, so that $250 / sensor starts to add up quickly. With all of Tesla's other sensors they probably wouldn't need 360 degree coverage - 90 degrees out the front would probably be sufficient, using cameras for side/back coverage.
    But, I think Quanergy is a great first step, and in 5 years they'll probably be an order of magnitude cheaper.
  • edited January 2017
    LIDAR is only necessary in the forward direction, which means one sensor only, just as there is only one forward facing radar currently.
Sign In or Register to comment.