'Self Driving Cars' aren't "Right around the corner"
The truth behind autonomous vehicles, and how companies like Argo AI are building a better automotive future.
Have you ever driven a car?
It isn't easy, and as humans, we're actually quite good at it. Piloting a 2-ton lump of metal down a road at 80 mph takes immense skill, and it's is easily one of the most complex things we as a species are capable of doing.
Yet, the current narrative surrounding autonomous cars would have you believe the opposite. An army of company PR men would gladly tell you that having humans on the road is a bad thing. From behind the thin veil of safety, they claim that autonomous vehicles will save us from ourselves and that we should help pay for our own replacement.
This is a bad idea.
I have yet to see compelling evidence confirming that "Self Driving Cars" are objectively safer than humans. Sure, a full suite of autonomous driving sensors can see far more than just one human driver, but it's a system's ability to react to information that would distinguish it from a human.
Many might argue that because nearly 100% of accidents are caused by humans, we should banish all humans from the roads. But isn't that the same as saying 100% of crimes are committed by humans, so we should get rid of all humans?
It's amazing how under the assumption that driving is easy, an entire industry has maligned consumers into accepting that replacing humans is a trivial inevitability.
And while most media outlets, online forums, and companies are guilty of perpetuating this narrative, it's hard to think of anyone at greater fault than Tesla.
Tesla's insistence upon calling their driver-assist system "Full Self Driving" has proven itself quite useful as a useful pr tool. And it's easy to find countless articles referring to Tesla's cars as "autonomous." A quick peek at the online configurator would suggest much of the same; it's only after you read the fine print below the $10,000 purchase button that the story becomes clearer.
Now, before I try and delve into why this naming scheme borders on criminal, I can't say that I feel entirely qualified.
I am fairly certain that 95% of people who write about "self-driving cars" don't have the slightest clue what they're talking about. And while I don't pretend to an expert by any means, I did at least have the sense to ask one.
Sam Abuelsamid is a Principal Analyst at Navigant Research who has spent his entire career working in and around the automotive industry.
And in response to Tesla's "Full Self Driving" feature, Sam wisely points out that "It is not self-driving, and it shouldn't be called full self-driving. Any system that requires a human to oversee it and be ready to take control at any time is not full self-driving, period. It's absolutely unconscionable and irresponsible for Tesla to call that system full self-driving at this point."
"Back in October of 2016, when they launched autopilot version two, on that conference call when Elon Musk announced it, the first sentence out of his mouth was that with software updates, this system has all the hardware necessary for level five full self-driving capability. And that was a lie. That was an absolute lie right from the very beginning. Because the definition of level five automated driving is that the system has to be capable of driving in all conditions all the time without any human supervision."
"They are trying to do a system with cameras and only a single radar sensor, which is never going to really be robust enough to drive everywhere under all conditions. You know, cameras are great for doing object detection and classification. But they're terrible for measuring distance. They don't work particularly well at night in a lot of conditions. They don't work in fog or rain very well. You need other types of sensors that give you additional information beyond what cameras can do to have a system that's reliable and robust enough to operate regularly without human supervision."
But thankfully, Tesla's false claims and misleading rhetoric have not appeared to quell the development of true AVs (autonomous vehicles). Serious companies such as Argo AI have spent years and billions of dollars on crucial research into what makes humans good at driving. Because the truth is that in the world of cars and technology, Tesla's first-mover advantage means almost nothing for the long-term.
The Wright Brothers were the first to take to the skies, but their minute-long flight didn't exactly guarantee them a spot alongside the likes of Boeing and Airbus. History has shown us that taking your time is often an invaluable business strategy.
And in speaking with Argo AI's Director of Communications, Alan Hall, I began to fully realize the importance of studying human drivers.
If you think about it, a tremendous amount of communication is done nonverbally. It's the same with driving. If someone wants to show that they're about to pull out of a driveway, they might inch forward, signaling that intent to other drivers.
"Our technology needs to fit in as the world is, versus expecting it to change for us." Says Alan, "And therefore, we expect to be operating amongst human drivers, pedestrians, bicyclists, scooter riders and other road users - and therefore prioritize our technology to drive with "naturalistic" intentions and behaviors, so it blends in with the local traffic, versus being an outlier."
'Full Self Driving' isn't "Right Around the Corner"
When discussing "Self Driving Cars," it's important to understand that there are echelons to their capabilities. Currently, systems such as Tesla's autopilot and GM's Super Cruise are somewhere in between "Level 2" and "Level 3" autonomy. Meaning that under the right conditions, the car can take turns, stay in lane, and is capable of managing certain aspects of driving, all while an attentive driver is ready to take over at any time.
"Level 5" autonomy is what most people mean when they say "Full Self Driving," and it is doubtful that we will see this come to fruition anytime soon.
This standard of autonomy insinuates that the car is capable of full independence under any circumstances, no matter where it is. Essentially, just like a human.
"Level 5" is what Elon Musk has championed for so long as "Right around the corner." But in speaking with multiple experts, it's easy to see that this is simply impossible.
It will take decades to develop, train, and test a system that can reliably maintain L5 autonomy.
Far more realistic is "Level 4" autonomy. This is what companies like Argo AI are chasing, and it's easily the best way to approach autonomous cars.
An L4 car will be capable of driving from point A to B without human involvement. However, unlike a Level 5 system, it will be geo-fenced to certain areas.
"We believe a clear understanding of the operating domain is also a critical factor in evaluating safety performance. It's important that there be clear rules of the road and safety metrics, so we're working with other industry stakeholders, as well as with federal, state, and city regulators and lawmakers, to develop these rules and metrics to ensure the safe deployment of self-driving vehicles." Says Argo AI CEO Bryan Salesky.
By acknowledging that Level 4 is the best path forward, Argo AI can focus its efforts on fine-tuning its system to operate in more populous areas. Thus allowing the technology to mature without being hampered down by the need to optimize their cars for the entire globe.
But crucially, Argo AI's mission "is not to replace the personal freedoms that driving provides, but rather to build technology that can help create compelling mobility products."