With Autonomous Cars, Small Detections Are A Huge Deal: Why Tesla Motors Inc. (NASDAQ:TSLA) ’s Autopilot is Fundamentally Flawed by Dr. Alex Lidow (full bio below)
In recent months, the public has learned of at least two crashes involving the Tesla Model S, which is the company’s luxury sedan that boasts of “Autopilot” capabilities designed to “make highway driving not only safer, but stress free.” In both crashes however, Tesla’s Autopilot feature was found to be at fault, including one on May 7 that killed the driver Joshua Brown. The National Transportation Safety Board (NTSB), which is conducting an investigation, released a preliminary report indicating that Autopilot’s driving assist system was in operation at the time of the crash.
Further, a few weeks later, Tesla admitted that neither Autopilot nor the driver noticed the white side of the tractor-trailer against a brightly lit sky. Last month, the company released an updated version of Autopilot, one it thinks will solve the problems that have vexed earlier versions of the software: rather than relying primarily on cameras and secondarily on radar, Autopilot 8 will lean mainly on the use of radar. As the company stated at the time of Autopilot 8’s release, “after careful consideration, we now believe it [radar] can be used as a primary control sensor without requiring the camera to confirm visual image recognition.”
Clearly a change was in order. However, there is still one major problem: radar is not the best choice technology in the context of autonomous or even semi-autonomous driving. And, it’s the inherent deficiencies of radar that I fear will lead to more Tesla S crashes in the future.
There are two key issues with radar: radar penetrates flesh, whereby human beings – adults and small children – not to mention dogs and other animals, appear semi-transparent. Secondly, radar only creates a “fuzzy” image of the surroundings much like a human with very poor vision. So in order to understand its surroundings, the vehicle must complement the radar with cameras. And in order to reconcile the two, computation is needed. And computation takes time. Which means there’s latency between when an object appears in front of a vehicle and when the vehicle actually recognizes it. Needless to say, all of this is problematic when speeding down the highway.
Put another way, radar doesn’t offer a direct view of a vehicle’s surroundings, but rather, an interpretation of what’s around it. And, that information can be interpreted erroneously.
So, the obvious question is, is there an alternative technology that doesn’t suffer from the same inherent limitations of radar? It turns out, there is.
LiDAR vs. Radar
LiDAR (light detecting and ranging, or light radar) is a technology that measures distance using laser light. Put simply, LiDAR systems fire laser beams in every direction and gauge the time it takes for the light to bounce back. By doing so, it creates an extremely precise 3D map of the car’s surroundings. The car then uses this information to determine what to do next. LiDAR is precise down to a resolution of just a couple of centimeters. Conversely, radar measures distance down to, at minimum, a few feet.
And then there’s the matter of speed. With virtually no computation required of LiDAR systems, latency is dramatically lower than with radar, allowing the system to create a detailed understanding of any obstacles in just a fraction of a millisecond. To put this in perspective, a car driving at 60 miles per hour travels just one inch in a millisecond.
Understanding all of this, are there any technological advantages where radar is superior to LiDAR? In a word, no. The myth of LiDAR not being able to see through rain and fog is just that. In fact, LiDAR sees through poor weather even better than people do. I don’t think we want vehicles on the road in weather conditions that are so poor that humans and LiDAR cannot see objects a few feet ahead.
And as noted earlier, the crash that killed Joshua Brown happened on a sunny day when the car’s Autopilot system couldn’t differentiate a white truck turning across the road in front of it from the bright sky behind it. LiDAR, unlike radar, can differentiate between a white truck and bright sky and would have prevented the fatal collision.
Where do Google, Ford and other big players stand?
While Tesla has opted to use radar, a number of large players in the autonomous car industry have already started to embrace LiDAR. Google is the most obvious example – anyone who has seen a Google self-driving car has noticed the LiDAR system on the roofs of these vehicles. Other big companies that have self-driving car projects involving LiDAR include Uber, Toyota, General Motors, Ford, Delphi, BMW, and Volkswagen, to name a few.
And, while the advantages of LiDAR are undeniable, the technology is only going to get better. In addition to the companies noted above, other organizations are making significant innovations in this area. For instance, MIT and the federal government’s Defense Advanced Research Projects Agency (DARPA) recently announced a collaboration in which they plan to squeeze an entire LiDAR system onto a single chip. My company, EPC, is another example: we make semiconductors, based on Gallium Nitride (GaN), that are hundreds of times faster than traditional silicon chips. All LiDAR systems on the road today use these chips in order to paint an extremely fast and accurate picture of a vehicle’s surroundings.
Where Is This All Going?
Despite LiDAR’s technological prowess, there’s no denying that, compared with radar, these systems have been too expensive. Until recently, LiDAR systems have cost as much as $75,000. However, a number of companies are rapidly bringing these costs down. For example, Quanergy Systems and Israeli startup, Innoviz Technologies, have pledged to introduce LiDAR systems that cost less than $250 by 2018. Moreover, the aforementioned effort by MIT and DARPA would enable LiDAR system chips to be manufactured in many commercial chip foundries, thereby lowering the cost of a LiDAR system dramatically, perhaps down to a few pennies per system.
Historically we have seen widespread adoption when technologies are both better and cheaper than the alternative. Which is why in another five to eight years if not sooner, I expect LiDAR, not radar to become as standard a technology as anti-lock brakes and airbags. If this claim sounds hard to believe, realize that many of the automotive innovations we now consider standard were at one time either prohibitively expensive or simply seemed in the realm of fantasy.
For instance, 36 years ago, I worked with General Motors, which told my team that it needed our help in devising a panic-stop capability that would actually prevent the wheels from completely locking up when drivers slammed on the brakes. This request sounded crazy at first. That is, until GM explained how anti-lock brakes worked. It then became obvious how such a safety mechanism could reduce accidents and save lives.
Given its high price point however – thousands of dollars per system – the technology didn’t immediately reach consumers. Rather, it