Tesla has done away with radar and is focusing on a pure camera sensor package for its driver assistance systems. The market researchers at IDTechEx think this decision is wrong.
The latest IDTechEx report, »Automotive Radar 2022 - 2042«, highlights numerous technological innovations that are driving radar performance into new, unprecedented realms. Why is Tesla, of all companies, foregoing radar despite all its advantages?
According to Tesla, radar occasionally leads to false readings, for example when a manhole cover is mistaken for an obstacle. This leads to a phenomenon known as phantom braking, in which emergency braking is triggered for no real reason.
Tesla is now training its neural network using good radar data so that the cameras can make the same depth and speed measurements as radars. Tesla says this is a success – at least under the right conditions.
When radar was first disabled, Tesla told its customers that there would be temporary restrictions on the use of their ADAS systems. Tesla limited auto-steer to speeds of less than 75 mph, increased the minimum distance to the vehicle in front, disabled emergency lane departure, and set the high beams to turn on automatically at night – presumably to compensate for the cameras' poor night vision. In addition, some customers reported diminished and poor performance in the rain. Not surprisingly, one of the main advantages of radar over cameras is that radar is hardly affected by poor light and visibility conditions.
Other manufacturers continue to rely on radar. According to the IDTechEx study, the number of radar units per vehicle will actually increase. This is due to the introduction of technologies such as blind spot detection and cross-traffic alert, which use radar to monitor the vehicle's blind spot environment.
It is also possible that radar will replace ultrasonic sensors, which are commonly used in parking assist systems today. This could increase the number of radar systems per vehicle to more than five. In addition, companies working on robotic axles are making extensive use of radar systems. Up to 21 radars per vehicle are used here.
Tesla pointed out in 2021 that driving through underpasses, for example, is difficult for radars because of their low height resolution. This was also true of the radar Tesla was using. The vehicle slowed down as a precaution because the radar did a poor job of detecting that there was clear space under the overpass. While the radar could be taught by software that a large signature such as that caused by an underpass should be ignored (since it can probably be driven through), this causes problems if there is a parked vehicle underneath.
Tesla used a Continental ARS4-B radar, which was one of the best radar systems in 2014. Since then, radar technology has advanced greatly. One measure of a radar's potential imaging performance is the number of virtual channels it has. This is the product of the number of transmit channels and the number of receive channels and is equivalent to the number of pixels in a camera. The Continental ARS4-B used by Tesla had 8 virtual channels (which was the norm in 2014). Since then, the industry has moved to 12 virtual channels. Continental's latest radars have 192 virtual channels. Startups like Arbe and Uhnder have more than 200 virtual channels, which could grow to more than 2,000 channels.
Part of the problem is the long life cycle of vehicles, which is typically 10 years. This is not unique to Tesla. If an automaker launches a new vehicle today and a breakthrough radar device comes out tomorrow, it will take up to 10 years for that radar device to be installed in the new vehicle. In other words, with any new vehicle nearing the end of its product cycle, the hardware is likely to be 5 to 10 years out of date, possibly even longer. Tesla's sensors were defined in 2016, so it will likely be 2026 before major changes are made to the hardware.
Tesla can counter this by making much of the vehicle software-defined. This allows the company to iteratively improve its products throughout the lifecycle through over-the-air updates. For camera-based systems, this works well because cameras produce a wealth of data and software improvements are still available to make the most of that data.
Si-CMOS radar only on the market since 2019.
The latest radars on the market and those currently being developed by startups are producing images that are much closer to LiDAR compared to the ambiguous scans of the past.
Part of this improvement is due to a transition in semiconductor technology. SiGe BiCMOS-based radars, such as the one used by Tesla, have been prevalent over the past decade. This is because they could produce a high signal-to-noise ratio compared to Si-CMOS-based radars. However, as the transistor size became smaller, Si-CMOS-based radars could match and even surpass the performance of BiCMOS. The advantage here is that the smaller transistor size allows more functionality and more virtual channels per radar. These Si-CMOS radars did not hit the market until 2019 and are not yet widely available. The current most powerful radars from startups are not even on the market yet.
According to IDTechEx, the new performance of radars could perhaps make Tesla re-evaluate its attitude toward radars.
IDTechEx is actively researching Mobility as a Service and autonomous driving. Its recently released report, »Automotive Radar 2022 - 2042«, is part of a broader mobility research portfolio that tracks the adoption of autonomous vehicles, electric vehicles, battery trends, and demand on land, sea, and air.