If you have a Tesla vehicle , you are interested in the following. In the following lines we talk about the decision of the popular brand to abandon the radar to focus on the cameras. IDTechEx experts have analyzed this situation, without understanding very well why Tesla has made this decision, given all the advantages that radar provides in vehicles.
Tesla claims that the radar occasionally gives a wrong measurement , for example mistaking a manhole cover for an obstacle. This results in a phenomenon called ‘ghost braking’, where the emergency braking response is activated for no real reason. Tesla’s approach is to use examples of good radar data to train its neural network so that cameras can make the same depth and speed measurements as radars. They say that this has been a success and that, under the right conditions, the technique can work very well. But what happens if the right conditions are not met?
When the radar was first disabled, Tesla let customers know that there would be some temporary limitations on the ability of its ADAS systems . Tesla limited the auto-steer function to less than 120 km/h, increased the minimum following distance, disabled emergency lane departure, and set the high beams to come on automatically at night (presumably to counter poor night vision). of the cameras).
Additionally, some customers reported reduced and poor performance in the rain. This highlights some of the main advantages that radar has over cameras. Unlike cameras, radar is not really affected by poor lighting and visibility conditions. The wavelength the radar operates on means it doesn’t see environmental occlusions like dust and rain particles, and because it emits a signal and looks for its own echo, it doesn’t matter if it’s day, night or even sunlight. direct.
In the many conversations that IDTechEx has had with major players in the automotive industry, it even seems likely that radar could start to replace ultrasonic sensors , which are often used in parking aid systems. In this way, the number of radars per vehicle could exceed five. In addition, the radar is a sensor widely used by companies that work in robotaxis, since some use up to 21 radars per vehicle. Therefore, if Tesla wants to set a trend, it does not seem that it will catch on.
Software moves faster than hardware
In a June 2021 Tesla AI director presentation, it was noted that situations such as traveling through underpasses are challenging for radars due to poor elevation resolution. This was true for the radar that Tesla was using.
The problem is that, with the poor elevation resolution, the radar finds it difficult to distinguish that there is free space under the overpass, so it will slow down as a precaution. Radar could be taught that a large signature, such as caused by an overpass, should be ignored (since it’s probably something you can drive through), however this creates problems if there is a vehicle parked underneath. The radar would still be unable to distinguish the overpass from the vehicle, a situation that could lead to a collision.
Tesla used a Continental ARS4-B radar, which was a perfectly good radar… back in 2014. Since then, radar technology has come a long way. 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 analogous 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). The industry has since moved to 12 virtual channels, but Continental’s latest radars have 192 virtual channels. Startups like Arbe and Uhnder and others covered in « Automotive Radar 2022-2042 » have more than 200 virtual channels, with room to grow to more than 2,000.
But this is not about blaming Tesla; many new vehicles are in the same boat. Part of the problem is the long life cycle of the vehicles, which is usually 10 years. This means that if a car manufacturer launches a new vehicle today and a game-changing radar comes out tomorrow, it will be up to 10 years before that radar can be put in the new vehicle. In other words, for any new vehicle nearing the end of its production run, the hardware on it is likely to be 5-10 years out of date, or possibly more. Tesla’s sensor suite was defined in 2016, so it’s likely that until 2026 there won’t be any big changes to the hardware.
Tesla can combat this by making much of the vehicle software-defined. This allows them to iteratively improve their products throughout their lifecycle via over-the-air updates. For camera-based systems, this works well, as cameras produce a large amount of data, and software enhancements continue to be available to make the most of that data.
The difference this makes to the imaging potential of a radar is enormous. The latest radars on the market, and those being developed by startups, produce much more LiDAR-like images compared to the ambiguous scans of the past.
Could Tesla change his mind?
Part of this improvement is due to a transition in semiconductor technologies. SiGe BiCMOS-based radars, like the one used by Tesla, have dominated for the past decade. This is because, compared to Si-CMOS based radars, they were capable of generating a high signal-to-noise ratio. However, by reducing the size of the transistors, Si-CMOS-based radars have been able to match and exceed the performance of BiCMOS. The advantage is that reducing the size of the transistors allows for greater functionality and more virtual channels per radar. These Si-CMOS radars did not enter the market until 2019 and are not yet in widespread use, and higher performance radars from emerging companies have yet to reach the market. However, the new heights of performance brought by radars could persuade Tesla to reassess its attitude toward radars.
Tesla often argues that humans drive with vision alone, so a vehicle should be able to do that, too. While this is true, it seems to be a bit of a closed mindset. Yes, we humans drive with only two eyes, but we don’t have many options. Tesla may be able to dispense with radar and go ahead with a vision-only approach. In IDTechEx’s view, this will hurt Tesla’s performance potential. When higher-performance radars come on the market, Tesla may rethink his decision.