So far the Autopilot is ‘just’ a driver assist system, but with version 10 it is supposed to become a fully autonomous driving system. The candle becomes the light bulb. Since October 2016, all Tesla Model S, X and now the Model 3 (so far 300,000 cars) are equipped with the Autopilot Hardware Kit 2, which consists of 8 cameras, radar, ultrasound and NVIDIA GPUs, but no LiDAR. Tesla’s stance seems to be that autonomous driving on Level 4 is possible without LiDAR. The majority of companies and experts developing self-driving cars today see LiDAR as crucial sensor, albeit a very expensive one. In fact there are startups that also try developing self-driving cars without LiDAR. Most prominent examples are AutoX and Comma.ai. Also, you’d have to think about the following: today’s Tesla’s have hardware built in, worth an estimated 2,000 dollars, which from a software perspective is not functional (and it’s not sure at all that the software will ever be ready). Try doing that in a traditional car company. But Tesla can, because Elon Musk! As simple as this! Since April 2017, Tesla owners could agree to allow Tesla accessing several gigabytes of driving data and videos every month. The data is then used for training Tesla’s machine learning systems to improve and develop the Autopilot and self-driving technology. Even if only a third of Tesla owners worldwide agreed to that, it’s still 100,000 vehicles. And such a number of vehicles, sending every month several petabytes of data to Tesla, including driving conditions in North America, Europe and Asia, might be a valuable motherlode of data. With the state of technology autonomous driving without LiDAR – which allows to build 3D models of the space and objects around the car – does not yet seem achievable, but – and that’s how the argument goes – humans as well don’t need LiDAR and can do that just with vision. The bet that Tesla sets on is that it’s faster with machine learning, and thanks to sensors used and sensor-fusion, to develop 3D vision just with that. Faster than the prices of LiDARs drop. It’s a race: are LiDAR prices dropping fast enough to get autonomous driving on the road with that technology, or is it the software approach without using a LiDAR that’s faster ready to deploy? If that succeeds – and it’s a big if – then Tesla would have in one over-the-air-update hundreds of thousands of fully autonomous cars on the roads. And that would motivate a number of companies betting today on LiDARs to abandon that technology and run to copy Tesla’s approach. But the verdict is still out. The request for beta-testers for the autonomous driving feature at Tesla may be a signal that Tesla has become confident that autonomous driving without LiDAR is possible. Only a signal, but potentially a glimpse of a technological tipping point. Here is a video how today’s Autopilot sees a city like Paris:
This article has also been published in German.