Cathie Wood, CEO, and Founder of a New York Based, ARK Invest shared a tweet that the reason behind why Tesla is now in favor of LiDAR with Pseudo-LiDAR.
In 2019, Elon Musk said, “LiDAR is a fool’s errand.” While every other company working towards autonomous driving system believe LiDAR has scope. Elon once said LiDAR won’t be used on Tesla vehicles. What happened? Was Tesla wrong for once?
They have been in the industry long, and they still plan to be. All this drama of not using LiDAR only lasted till Tesla made an in-between option, Pseudo LiDAR.
Avanced in LiDAR by Tesla
In February, Andrej Karpathy, senior director of AI, presented Tesla’s methods of AI during a scaled ML conference. Tesla’s progress towards the autonomous driving system is ahead of the other cars, even before choosing the Pseudo-LiDAR.
One of the main reasons why Elon Musk was against the idea of using LiDAR technology was because it was costly. Now, Tesla successfully made a much cheaper option by imbibing its AI technology with LiDAR. Comparatively, this is a cheaper model.
As said by Karpathy, the advancing process of Tesla’s AI doesn’t rely on cameras but relies on neural nets and advanced processing. This kind of focus on neural nets allows the system makes sense of the surrounding area. Eventually, making it more powerful with linked with LiDAR cameras.
How is Pseudo-LiDAR different?
Traditional LiDAR technology is something like sending out lasers all around the car. And using these signals to identify and analyze the available data. Tesla argued that their technology should be advanced, not the camera. As one laser costs at least a thousand dollars, having eight such laser emitting devices on a single car is costly and not practical for massive sales in the future.
Instead, the advanced technology by Tesla can now identify the objects from a video. Meaning, the system can take the video from a camera and try to make sense of the objects. This pseudo-LiDAR doesn’t have the need to use lasers. Thus reducing the costs by thousands of dollars.
— Cathie Wood (@CathieDWood) May 1, 2021
As tweeted by Facebook CTO, Mike Schroepfer, random data (random video) can be sent to the technology, and the system detects the objects accordingly. This advanced technology can detect objects, categorize them into groups and tell under what category the objects are from. As shown in the above tweet, it would say, “animals”, very precise than just saying “living object”.
Furthermore, they even extended this technology with art videos too. Their system was able to identify the art names, as shown in the tweet thread by Mike Schroepfer.
What’s happening here is one of the defining AI advances of recent years — self-supervised learning — enabling new systems to develop much deeper and more generalized understandings of the data they’re trained on.
— Mike Schroepfer (@schrep) April 30, 2021
How far the companies working with LiDAR go?
By last year, Google’s Waymo, GM’s cruise, and other AI technology successfully acquired LiDAR technology. Even one of the world’s largest chip maker, Intel has been working on creating their own LiDAR technology.
Among them, Google’s Waymo was successful enough to make it commercially available. Additionally, Waymo’s LiDAR is known as Laser Bear HoneyComb.
All of the details were fine, except there was no information about the price and scale of availability. Also, HoneyComb LiDAR is mostly based on mechanical technology. They added the feature of mirrors and rotatable cameras (turntable).
These spinning radars where the laser beam can rotate too, resulting in using lesser lessors could be beneficial. However, the learning scope for this technology hasn’t progressed and is far away from being able to detect objects.