Autonomous Vehicle Tradeoffs
It seems like there’s a new story about autonomous vehicles in the news almost every day, and one day’s message flat-out contradicts whatever you heard the day before. One day, self-driving cars are taking over the streets; the next, they won’t be a viable technology for another ten years. One day, cameras have been overtaken by lidar as the best sensors; the next cameras are replacing lidar as the only feasible alternative. One day, autonomy is going to eliminate the jobs of millions of drivers; the next, it’s something drivers of today have no cause to worry about.
So, how do you know what’s really going on in the autonomous vehicle industry?
One obvious source of confusion stems from there being multiple viewpoints of a complex industry. One analyst tries to read the tea leaves and comes to her own conclusion; another does the same and arrives at a completely different view. Meanwhile, many startups are doing their best to hype the industry because they stand to profit from investments, while others are sounding dire warnings because their interests are closer to those of a competing industry or interest group.
In wading through the rival messages about self-driving technology, it also helps to keep an idea in mind that was best expressed by the economist Thomas Sowell. “There are no solutions,” he says. “There are only tradeoffs; and you try to get the best tradeoff you can get. That’s all you can hope for.”
Tesla’s Big Self-Driving Tradeoff
Tesla is a good place to start because the company, or often its president Elon Musk, is in the news about every day. Early this year, Musk announced the release of a beta version of “Full Self-Driving” to select owners of Tesla cars. Now, when someone says, “Full Self-Driving,” it definitely sounds like you’re saying the thing drives itself. The thing is, the company uses this term when discussing the system in public, but when they talk to regulators behind the scenes, they admit they’ve really only achieved Level 2 autonomy, which means drivers must not only stay alert but should really keep their hands on the wheel.
In other words, Tesla’s full self-driving mode is really just an advanced driver assistance system (ADAS). What’s going on here is that the company has to get as many of its vehicles on the road as possible because its self-driving system relies on machine-learning technology—the more miles traveled the better the system gets. Tesla obviously also needs to sell cars to stay solvent. So, the tradeoff that comes with overhyping their product has them getting lots of cars on the road while also facing a bunch of criticism and pushback about their claims.
Waymo is on the other side of this tradeoff. Their taxis operate at a higher level of autonomy—probably Level 4, which is fully autonomous but with a backup driver—but they have a limited range. Instead of machine-learning programs that rely on camera footage and miles driven, these taxis use lidar and highly detailed maps of the routes they routinely take. Not surprisingly, Waymo representatives are happy to call out Tesla for its exaggerated claims.
Tesla’s approach may eventually lead to vehicles that achieve the higher levels of autonomy with their already far more expansive range. Waymo may be able to expand the range of their vehicles by creating more maps of different regions. Or both approaches could pay off, and the spoils will go to whichever company reaches the finish line first.
Waymo is also betting on lidar as opposed to cameras, taking the opposite approach to Tesla’s. Musk has famously said, “Anyone relying on lidar is doomed.” Indeed, lidar is much more expensive, draws far more power, and is easily tripped up by rain and snow. That’s why Waymo’s vehicles are restricted to places like Arizona, where the sun shines nearly all year. It’s also hard for a technology that uses time intervals between pulses of light and their reflection to read street signs or parse traffic signals.
There’s another potential cost to lidar that will become more pronounced as more self-driving vehicles start operating on the roads. The light pulses have to avoid certain wavelengths that can cause damage to the human eye. The adjustment seemed relatively simple at first, but the new wavelengths are now posing another threat—this time to cameras. That means any vehicles relying on cameras for navigation—like Teslas—may be at risk. Avoiding the wavelengths that are dangerous to human eyes also limits the technology’s range of effectiveness.
Autonomous vehicle engineers can turn to cameras, but these come with tradeoffs of their own. Lidar measures distance actively and directly, generating data rapidly and reliably. To gauge distance with cameras, you have to process overlapping images from different vantages, which means multiple cameras and added lag time for computing.
Cameras also tend to be less than perfect in rainy or snowy weather, a challenge engineers may overcome by installing entire arrays of cameras and hooking them into systems that increase resolution through data redundancies. But that solution adds layers of complexity, which means more ways for things to go wrong.
Safer Roads but Fewer Jobs?
Whenever you mention the coming age of autonomy, most people’s minds go right to the tradeoff between jobs and safety. Entrepreneur Andrew Yang rose to prominence in his presidential campaign of 2020 based on his warning that automation would eliminate countless jobs in the near future, most immediately perhaps in the transportation industry. But, given the number of accidents caused by distracted driving and other human errors, might this scenario still be worth it?
Waymo recently conducted virtual reenactments of all the fatal accidents in Chandler, a town in the Phoenix area, for the past ten years and found that their self-driving systems would have avoided nearly all of them. The system even avoided 80% of the accidents caused by other drivers. These recreations were simulated of course, but that’s an excellent safety record nonetheless. So should we accept the reduction in jobs in light of such a big reduction in severe accidents?
Fortunately, in this one instance, it turns out that the tradeoff probably won’t be as costly as most people think. A recent modeling study of the economic impact of autonomous vehicle adoption by the Department of Transportation found that the overall GDP will likely increase significantly, even as there will be fewer layoffs of truckers than earlier analysts assumed. This is because the transition to autonomy is happening more gradually than many anticipated, and there’s already a major shortage of drivers today.
So what’s going on with the autonomous vehicle industry?
Companies are making bets on different sides of various tradeoffs, as we’ve examined here. But another thing to keep in mind is the tradeoff between testing self-driving systems on the roads—higher payoffs, bigger challenges—vs in places like farms and mines (and even in the air), where the stakes are lower. Of course, it’s the roads we’re most interested in. And it’s probably a safe bet that, Musk’s promises notwithstanding, the majority of vehicles on the road are going to be less than fully autonomous for the rest of the decade. But that doesn’t mean the next nine or so years aren’t going to give us some exciting developments and big surprises.
Also Check Out:
From Cameras to Lidar and on to Camera Arrays for Autonomous Vehicles
Light Fields and Micro-Lens Arrays for Waymo's Autonomous Vehicles?
Tesla’s Cameras vs Everybody Else’s Lidar
Follow us on:
You can also sign up for our newsletter, so you never miss a post. Just scroll down to the form in the footer.