Risky or brilliant? Tesla Vision

As of April 27, 2021, radars will apparently no longer be installed in the US versions of the Model 3 and Y, Tesla announced. On the support page, the company published a statement, which at the same time introduced Tesla Vision as the new basis for Autopilot and, consequently, the upcoming Full Self Driving (FSD).

We are continuing the transition to Tesla Vision, our camera-based Autopilot system. Beginning with deliveries in May 2021, Model 3 and Model Y vehicles built for the North American market will no longer be equipped with radar. Instead, these will be the first Tesla vehicles to rely on camera vision and neural net processing to deliver Autopilot, Full-Self Driving and certain active safety features. Customers who ordered before May 2021 and will receive a car with Tesla Vision will be notified of the change in their Tesla Account prior to delivery.

For a short period during this transition, cars with Tesla Vision may be delivered with some features temporarily limited or inactive, including:

– Autosteer will be limited to a maximum speed of 75 mph and a longer minimum following distance.

– Smart Summon (if equipped) and Emergency Lane Departure Avoidance may be disabled at delivery.

In the weeks ahead, we’ll start restoring these features via a series of over-the-air software updates. All other available Autopilot and Full Self-Driving features will be active at delivery, depending on order configuration.

All new Model S and Model X, as well as all vehicles built for markets outside of North America, will continue to be equipped with radar and will have radar-supported Autopilot functionality until we determine the appropriate time to transition those vehicles to Tesla Vision.

The text also explains the current limitations that are not considered tested by the U.S. National Highway Traffic Safety Administration (NHTSA) and are therefore temporarily disabled or have limited use. These include forward collision warning, automatic emergency braking and lane departure warning / avoidance.

Why No More Radar?

So what could be the reason that led Tesla to leave radar out of the vehicles completely? And what does that mean for Autopilot and the FSD?

First of all, it is important to remember that Tesla CEO Elon Musk has repeatedly spoken out against another type of sensor, namely LiDAR. He referred to LiDAR as a “crutch.” While most other companies developing autonomous driving systems consider LiDAR an essential part of the sensor suite, Tesla has been shipping its vehicles with 8 cameras, radar and 2 AI chips for autonomous driving for three years. This so-called Hardware Kit 2.x is intended to provide the technical basis for all Teslas equipped with it – currently more than 1.5 million vehicles – to use FSD and thus achieve Level 4 autonomy.

Although sightings of Tesla-owned Teslas equipped with Luminar LiDARs are being shared on social media, Robocar expert Brad Templeton argues that this is not a 180-degree turn by Tesla and an indication of a future use of LiDARs, but rather control runs comparing the accuracy of the cameras with LiDAR data, and then in Tesla’s machine learning system to help improve the purely camera-basedAlgorithms.

Brad Templeton now speculated that the chip shortage may have first encouraged Tesla to omit the radars as well and proceed with the camera system only. It has been reported in the media that Tesla has produced more than 10,000 vehicles in stockpile and cannot deliver them because a key component would be missing.

Omitting the radar would only be consistent and would offer several advantages. First of all, production is cheaper and components that are susceptible to failure or expensive to replace in the event of accidents are eliminated. Then a machine learning system no longer has to make decisions about which sensor to trust more when the different sensor modalities don’t agree on what they have in front of them.

However, it is unclear whether cameras alone will ever be enough to drive autonomously. Critics point to the resolution of the cameras in the Teslas, which they say are insufficient. Also, pure camera-based control struggles with fog, dust clouds, rain and snow, just as humans do. Others have contrasted current LiDAR technology and the current cameras with the corresponding algorithms and compared the result. This results in discrepancies that might not be significant in some cases, but can make the difference between a collision or safe driving in others.

In the short term, therefore, many cannot imagine that comparable performance of a purely camera-based self-driving technology is possible, as is the case with a multimodal sensor suite. However, the past has also shown that algorithms always win in the end.

What are the possible consequences?

Tesla CEO is known for the art of the big bet. Rockets that land again. Electric cars that are now displacing internal combustion vehicles, payment systems that rival banks and currencies, and the many other small innovations at Tesla that the competition still hasn’t figured out or is aware of. The list of his winning bets is large. He didn’t always win them in the time frame he stated in his master plans, he delivered every time.

If Tesla succeeds in operating the FSD with cameras only, it would be a game changer. The cards would be completely reshuffled. Several million Teslas would then be capable of driving autonomously, and this would on the one hand call into question the strategies of all other manufacturers who have so far relied on expensive LiDARs. Tesla would not only be ready for use in a few cities with a fleet of robot taxis, as is to be expected from the other suppliers, but Teslas would be usable in all cities and regions at a stroke, without Tesla having to go to gigantic expense to buy fleets and equip them with the self-driving technology, as Waymo, Zoox, GM Cruise and others are planning. The deployable Teslas would all be paid for by the customers themselves, maintained at their expense, and the supply can respond flexibly to demand. And Tesla would centrally coordinate the vehicles and provide the cab service, at prices that can far undercut everyone else. Uber, Lyft, cabs, public transit, and even Waymo would be disrupted in one fell swoop.

The cost comparison with other providers would look as follows:

Costs per Mile in Cents

The risk Tesla is taking here is huge. The chance that Tesla Vision will never work with the certainty needed for mass deployment is great. But the odds, if it does work, are even greater.

Timeline

What schedule can be expected? After Elon Musk has already announced several times that the FSD beta, which is currently being tested by 2,000 customers, will be made available to all customers in the next few weeks clearly as a level 2 version with appropriate caution, it can be expected – despite the delays – that it will be made available in the USA in the 2nd half of 2021. For at least the next 1-2 years, it will then run on an estimated 100,000+ vehicles in test mode. Data will then continuously flow back to Tesla into the FSD machine learning system until it is safe enough and Tesla will seek official approval and it will be tested by regulators. Approval could happen around 2025.

TimelineNumber of robotaxi fleets in cities / regions2021up to 3 in the USA20225 to 10 in the USA202525 to 50 in the USA2023-2025first fleet in EuropePossible number of robotaxi fleets (without Tesla)

Compare that to a possible rollout schedule of other manufacturers’ robotaxi fleets with LiDAR, camera, and radar-based systems, and we could see that number of robotaxi fleets in cities and regions. This does not even take into account possible fleets in Asia.

We won’t know how it will turn out for a few years, but it could turn out to be Elon Musk’s most brilliant move, or the dumbest and riskiest. But as I said, Elon Musk is one of the last exponents of the Art of the Big Bet.

This article was also published in German.

Read More