Elon Musk has cast doubt on Google's choice to use LIDAR sensors in their self-driving car program. A comparison of Google and Tesla's approaches to autonomous vehicles may explain why.

While Tesla's CEO, Elon Musk, and Google's co-founder, Larry Page, are friends, Musk has criticized Google's use of LIDAR ((LIght Detection And Ranging) in their early-stage self-driving cars. 

At a press conference held last October, Musk claimed that the LIDAR technology "doesn't make sense" to implement in an autonomous car and that he's "not a big fan of LIDAR". While he doesn't state directly that Google is using the wrong technology for their cars, he does believe that it isn't something Tesla will implement in their Autopilot systems. 

Regardless, with his prenotions of using LIDAR in self-driving cars, Musk utilized LIDAR on SpaceX's Dragon to determine the range of the spacecraft whilst docking to the ISS. 

Musk has made a name for himself by creating and popularizing ideas such as PayPal, SpaceX, and Tesla. However, he might be wrong about Google's approach to creating the first self-driving car. Google's Self-Driving Car Project was created to ensure a safe mode of transportation that would completely eliminate any human error. Google has taught their cars to navigate through city streets using sensors and software that sense and avoid objects such as animals, cyclists, pedestrians, vehicles, and more.

Below is a short video explaining how a self-driving car performs throughout the city:


Google's Use of LIDAR

While Google and Tesla have a common goal of making transportation on the road safer and eventually operator-free, their techniques are quite different.

Google self-driving cars use LIDAR, which maps out the car's surroundings using lasers. LIDAR measures the shape and contour of the ground from the sky. It reflects multiple laser pulses off of objects that surround the car and measures the distance and time that each pulse has traveled.

From these measurements, the LIDAR system can provide accurate information on the height and distance of objects.


Image courtesy of Google


While this system is taking great strides in creating a fully driverless car, it comes at a hefty cost. Google used Velodyne's 64 Channel LiDAR sensor, whic is priced at about $80,000 for just one sensor.

This might be one reason that Elon Musk refuses to use LIDAR technology in Tesla's Autopilot systems.


Tesla's Use of Optical Sensors and RADAR

Musk believes that passive optical sensors and a RADAR system can accomplish the same thing as Google's LIDAR system.

Musk equipped Tesla vehicles with 12 long-range ultrasonic sensors that provide a 360-degree view around the vehicle. In addition, each vehicle has one forward-facing RADAR system. Integrating these components together helps power Tesla's Autopilot system.

Like a LIDAR system, a RADAR system sends out signals, but in the form of periodic radio waves that bounce off of objects in the cars proximity. Once they hit an object and return to the car's system, it will measure the time it took for the radio waves to travel to and from the object.

The advantage of radio waves is that they can be transmitted through rain, snow, fog, and even dust. By comparison, laser beams used in LIDAR cannot.


Image courtesy of Tesla


Different Tools for Different Tasks

While Elon Musk has made some bold comments regarding Google's methods, most people don't realize that they are working on two completely different tasks. Google is using Velodyte's 64-channel (64 beams) LIDAR to place itself with an accuracy of 10 cm on a pre-existing map. Google is also utilizing the LIDAR system to not only create a 360-degree model of the car's surroundings but also predict the movements of nearby pedestrians and vehicles.

On the other hand, Tesla's Autopilot system uses forward-facing cameras that are produced by Mobileye. These cameras can accurately pick up the location as well as curvature of highway lane markers which help keep the vehicle in its lane and make basic lane changes.

While each company's technology is exceptional, they each work towards different results. Tesla's Autopilot system is inexpensive and will prove to be useful for Elon Musk's initial goal: automating 90% of driving within just a couple of years. However, the other 10% of driving scenarios are quite hard to implement.

Google began working towards this same goal quite some time ago and instead focus now on something different: a fully autonomous car that will completely eliminate any human errors. 

So, was Elon in the wrong for calling out Google on their LIDAR system? Well, yes and no. It is a pricey system that will eventually need to be altered if Google wants to make it affordable for consumers. However, Google is working on eliminating the need to drive a car while Tesla is working on providing for eliminating some of the daily driving we need to do. From that perspective, LIDAR might be the right choice after all.




  • SalceyForest 2016-08-01

    Occurs to me that a fully self-driving car would have its seats arranged as a mini railway carriage so that occupants could communicate better. My question is: Would you want to sit with your back to the direction of travel? I wouldn’t but maybe I’m old!

  • dirtmover 2016-08-05

    “Tesla’s Autopilot system is inexpensive and will prove to be useful for Elon Musk’s initial goal: automating 90% of driving”

    The problem is the more you automate and the better these systems get the less attentive the operator becomes. Musk will reach a point, and recent events suggest that he may already have got there, where (some) drivers may lose sight of the fact that the vehicle is not 100% autonomous and start doing other stuff while not paying attention to what the vehicle is doing. When the system fails on something it is not programmed for or simply does not have the feedback to manage the driver is not ready to take over.

    These systems need to be either 100% autonomous or pretty dumb and leave the driver 90%+ in control e.g. simple speed control and/or electronic assistance for braking, stability, parking etc. IMO this middle ground that Musk is trying to play in is a recipe for disaster.

  • ronsoy2 2016-08-05

    Long way off. It would work in an isolated city where ALL cars are controlled by a central computer but in an open city with traffic, not! Note that in the test video there was no traffic! And no other cars with the same system bouncing their detection beams randomly around! And not in the rain. And with LOTS of electronics that all has to be at least medical qualified if not military qualified reliability. (if not, are you going to put your kids in it and send it to school? HA!)

  • Best_Intentions 2016-08-05

    I am a baby boomer (over 60) so my opinion has to be viewed from that stand point.  Personally I love the idea of a fully autonomous vehicle.  But the direction of development is trending towards a status quo where the driver will be expected to take over in an emergency.  At this point in time normal (human controlled) driving requires a form of constant vigilance in order to get from point A to point B.  With the advent of autonomous vehicles this talent/skill of observing and reacting quickly in an emergency driving situation will not be exercised and, as a result, atrophy.  Come the time that you are required to interpret and intervene in an emergency (i.e. potential accident) situation the driver is likely to find that even though there was enough time to avert the accident their skill set was just not up to the task.

  • originalgeek 2017-03-10

    Good comments and good article (for a start).  Computer Vision approaches get blinded by weather, Ultrasonic sensors do not have the desired range (the “long range” descriptor is compared to traditional ultrasonic sensors which are in the <5 meter range) aren’t precise the further the range.  LIDAR penetrates adverse weather the best.  As for the cost, you’ve quoted the oldest model.  There are many companies (including Velodyne) that are coming out with low cost (sub $1000 easily, probably sub-$500) in the current time horizon.  A note on Level 3 vs. Level 4 autonomy.  The commenters are right-on. (Then named) Google proved exactly the point that relying on a human to intervene “when needed” doesn’t work.  Our reaction times are horrible in any case.  And that was with paid testers in the driver’s seat! How much warning does an ADAS system have to give a driver and would it be effective?  I’m guess that time is measured in seconds, not milliseconds.  Finally, there’s no mention of HD Maps (that are up-to-date).  Both MobilEye and Waymo are counting on these, no mention of what Tesla is doing.  Being a pilot, I’ll also use that history as analogies for ADAS:  The pilot/automation hand-off is where many accidents happen (AF447, AA965, West Air Sweden 294) and usually pilots have more time than we would in a car to figure things our.  On the HD Map portion, navigation maps are updated on a 28-day cycle, ADAS maps probably have to updated within days.  Just some food for thought.  (keep it up! This is all new)