Autonomous driving tech has been ‘almost here’ for years now, but why and where is it stuck? And how soon can we expect full self-driving cars.
For all the publicity, all the videos of Teslas driving themselves and all the talking up, Teslas are only Level 2 autonomous—to put that into perspective, SAE International (the arbiter of all automotive standards) defines six levels of autonomous driving capability numbered from 0 (no autonomous ability) to 5 (full self-driving). In truth, quite a few automotive manufacturers are on par with Tesla, at least when it comes to the level of self-driving capability that their cars have. There are a few that will have Level 3 Autonomous Driving capability built in shortly. Some self-driving pioneers like Waymo (with autonomous tech outfitted Jaguars in their fleet) are skipping Level 3 entirely and aiming for Level 4 (full self-driving capability in a geofenced region). So, most of Elon Musk’s and Tesla’s hyped-up claims of full self-driving are just that—hype. Truth is that Tesla, with its decision to eschew lidar, is not in a position to be able to bring full self-driving to the masses anytime soon. In fact, Tesla’s full self-driving beta has been caught wanting in scenarios Waymo robotaxis have been handling for years.
So then who is actually the leader when it comes to self-driving tech? Alphabet’s Waymo is clearly at the top by all accounts. It is the only company with a fully functional self-driving service on the roads. Waymo robotaxis have been plying on the streets of suburban Phoenix, Arizona, for a few years now. In fact, such is the confidence they have in their system that these robotaxis can operate without a human-safety driver. A select group of vetted individuals can summon a robotaxi anytime in low pedestrian traffic in the suburbs of Phoenix.
Then there’s General Motors’ Cruise, which plies on the streets of San Francisco. Both Waymo and Cruise are aiming for Level 4 Autonomous Driving capability. When it comes to distance driven autonomously, Waymo is the clear leader with well over 20 million miles. Another metric worth considering is the number of times that human intervention is necessary—“disengagements” in autonomous driving speak. Waymo claims that for every 1000 miles driven, its robotaxis only has a corresponding disengagement figure of 0.09. This means that human intervention is only necessary once every 11000 miles. It has to be said, however, that the wide, well-marked streets with low pedestrian traffic in Phoenix, Arizona, aren’t representative of other urban centers around the world.
Waymo did make bold claims about there being far more of its self-driving robotaxis on the roads by now—much more (to the tune of 82,000). Today, by all measures only a few hundred are in actual service. Waymo is not the only company which made huge claims about how imminent self-driving cars would be. Tesla would be the biggest culprit. Elon Musk had claimed that full self-driving cars would be available in 2017 and quite a few times since then. Tesla now claims that its cars will offer full, self-driving by the end of the year. There has been, as of writing this piece, no factual evidence that corroborates that claim. Worryingly, Tesla’s use of the term “Full Self-Driving” when mostly referring to highway driving with a human driver firmly behind the wheel, is both extremely irresponsible as well as dangerous. Tesla’s cars using Autopilot have been involved in quite a few fatal accidents in the recent past, and there is a massive federal investigation in the US about Tesla’s Autopilot that is currently underway. Regardless of what transpires in due course, one thing is for sure—misleading claims about the abilities of its cars could have terrible consequences as drivers take these cars to be more capable than they really are. In time, it could also make the public grow wary about safety when it comes to self-driving cars.
Over the years, the self-driving tech space has consolidated with several big names either shutting shop or getting acquired. Ride-hailing startups like Uber and Lyft, who could be the biggest beneficiaries of self-driving tech when available, have given up on their autonomous driving dreams. The amount of research and engineering that has gone into self-driving cars has been incredible. And so has been the amount of money that some of the world’s tech companies have spent. But it is now clear that the timelines that these companies promised were too optimistic at the very least. Artificial Intelligence is at the core of self-driving tech. Sure, the cameras, the sensors, and the detailed maps are bare necessities but AI is where the magic happens.
The 2010s saw an exponential growth in this space, and the advancements that were realized prompted most self-driving pioneers to make bold predictions and subsequently pour billions into them. The AI improvements that most of them were banking on never arrived. Even with tens of millions of miles of training data, it is hard to train AI systems to react to the odd curveball that normal driving throws up. Getting enough training footage and data of accidents, for instance, is proving to be extremely difficult. Human beings by their very nature are adept at tackling a problem that requires instinct and understanding what other human beings could possibly do. Take, for instance, a roundabout. Human beings can identify body language cues and easily navigate a roundabout; self-driving cars can’t. Things as common as a traffic cone or even a setting sun have been known to confuse the self-driving cars of today.
Self-driving tech has definitely made massive improvements and are now safer than ever, but they are still nowhere as close to how safe human drivers are on an average. Humans have an average of one fatal collision every 100 million miles driven. With Waymo only having done 20 million miles and its rivals significantly less, we have already had several fatal collisions where self-driving tech or its non-ideal implementation is partly to blame (Waymo, however, still has had no fatal accidents).
So self-driving cars haven’t been as safe as human drivers and aren’t yet equipped to handle our cities. Does that mean our dream of being driven around by AI is just a dream? The obvious opportunities that self-driving can open up are far too great for tech companies to abandon it, despite the past few years proving that making real headway is far too difficult an engineering problem than previously envisioned. But perhaps, self-driving would instead make its way first to commercial buses and long-haul trucking—two sectors with low route variance. Waymo, Aurora, and Nuro with its acquisition of Ike are hedging their bets here. Or maybe Nuro with its fleet of driverless, low-speed delivery robots could crack it first. It’s hard to say at this point.
One thing that we now know for sure, however, is that self-driving companies have settled in for the long haul. Full Self-Driving by their own admission is possibly quite a few years away. Chris Urmson, chief executive at Aurora, the company that swallowed Uber’s self-driving division, reckons that it will be a gradual transition that could even take 30 years. There are some who are far more optimistic. But that optimism alone will not translate into quicker deployment of self-driving technology. For that to happen, breakthroughs in AI have to happen like they did in the 2010s. AI-based systems could not identify pictures of pets just ten years ago, and can now create albums and slideshows of your pets with no intervention from you. To bet against a game changing, new innovation in AI in the next few years would be rather foolish, especially with the amount of investment being directed to it.
In short, all the talk about us commuting in self-driving cars by now was hype that was pushed out by companies with an active interest in getting people to believe that self-driving cars were just around the corner. The dramatic improvements in AI also played a part in making us believe that they were in fact just a few short years away. Now, it is obvious that cracking the self-driving code is not a sprint, but a marathon. When would we actually see self-driving cars? Well, my guess is as good as yours. But one thing is for certain: The developments in the space have brought us much better Advanced Driver Assistance Systems (ADAS) that help you steer, keep a set distance to other traffic, and brake autonomously. For the most part, it is these ADAS systems that we will see put to great use in making driving less cumbersome in the near future than it is right now.