Why the Sleek Shall Inherit an Autonomous Earth (first)
So far, it has always been about making existing car models autonomous. But what about ‘reformatting’ the car before the best available AV technology goes in? In the land of the blind, the one-eyed man is king. In narrow streets, it seems obvious that narrow vehicles do a better job at self-driving, as illustrated in the following clip by Argo AI: https://youtube.com/watch?v=RzNXnYqkV7w Below: the equation that brings together two ambitions — ‘greening’ the car and making it driverless are two sides to the same coin.
SEVEN solid reasons why SAE Level 5 Autonomous will take longer to realize, and which I haven’t seen AV developers address properly.
1.Mimicking the human brain. Billions of brain cells multiplied by tens of thousands of synapses in each individual human brain make for more instant connections than there are stars in 20 to 40 thousand galaxies. Housed in a roundish ‘cockpit’ capable of swiveling (human head), equipped with two amazingly effective optic and ditto hearing sensors (eyes and ears) and a hard-drive full of constantly updatable tutorials, topped off by the human intuition, they enable us to split-sec reference what we see, hear and feel when driving. Hard to beat those, if ever.
2. The Element of Risk. The awareness of danger is something that can’t be synthesized by an AI that easily. To assess and differentiate different forms and levels of risk and danger is like a second nature (developed through thousands of years of human evolution), that flashes through the human brain within milliseconds. For an AV it should not be any different, yet it’s hard to learn or mimic. Below: ‘high-tech’ is steering (pun intended) us towards distracted driving OR to autonomous — BOTH are responsible for substandard car control. And I haven’t even discussed the tricky subject who to blame when an AV becomes involved in an accident: the human behind the steering wheel, the carmaker, the AV tech supplier and/or the software developer?
3. The Simulation ‘Game’. Developing a simulation with 100% accuracy might require as much problem-solving and resources as developing self-driving technology itself. It is hard to switch from ‘self-learning’ the AV to fail-safe self-driving for real.
4. ‘Other cars’. Ever realized that before you can expect driverless vehicles to take over 24/7, you will probably need to phase out conventional, manually operated cars, and will probably need overlaying grid control? “The only way to effectively address street capacity problems with autonomous vehicles is to switch to 100 percent autonomous driving.” For now, AVs will be limited to taxiing in cities with wide enough roads (Waymo in Phoenix), on more or less designated lanes, during traffic-low nighttime slots (GM Cruise in San Francisco), to campuses, industrial parks, airports, gated communities, or the type of ‘smart cities’ like the one Toyota is working on right now.
5. ‘Road anomalies’. A lot will probably need to happen before AVs can be considered safe. Intersections may need to be changed, improved, standardized. ‘Road anomalies’ may have to be removed, not to confuse an AV. It’s going to be a long and very costly process. Until that time, slow and cautiously maneuvering AVs will actually contribute to more congestion.
6. No Standardization in sight. There’s a winner-beats-all mindset, in which each AV developer is working on its own to bring an AV. Not good from a standardization perspective, which is the most important prerequisite for any system that’s supposed to take over from human drivers in a regulated and controlled manner. That requires all developers to open up and share their proprietary secrets. Fat chance that will happen any time soon. There is less tolerance towards failing autonomous driving systems than towards ten (hundred?) times as many human drivers who cause road accidents.
7. The displacement aspect. The types of SUVs and MPVs used by ride-hail providers, are not conducive to autonomous maneuvering, let alone make a profit. Vulnerable road users like pedestrians and cyclists should of course not become casualties of robo-taxis that can’t quite cope with traffic, because of its inherently complicated nature and unpredictable fluidity. Self-driving, or any driving that’s taking place, is first of all a matter of using available road space, and therefore vehicle size. Imagine twenty physically challenged (you’re not supposed to say ‘obese’ any more) walking down a narrow stretch of pavement. Now imagine the same number of athletes walking on the same pavement… The smaller the AVs, the more margin to maneuver and evade other road users.
Did you know that, according to WSJ and MIT estimates, so far $100 billion has been spent on AV development. Fortune already warned for the fallout when predictions and pitches fall short of expectations. IMO, there is sort of a grey zone between “Fake It Till You Make It” that led to the downfall of an individual company like Theranos… and the type of wishful engineering that is collectively hyped by a bunch of companies. Hyperloop is a good example of this.
Will AV development turn out the same? Investors better do their homework (due diligence). Below: Waymo was on to something when it introduced its ‘helmet-on-wheels’ a decade ago. Rounded, sloping contours sharply reduce fragmented scanning and imaging. Now make it more comfy and fun, safer and have it accommodate three. You end up having a car that can be used as a Next-Gen rickshaw as well as an AV for personal use and pleasure.
Personally, I think that SAE Level 5 autonomy of passenger cars will take a long time to realize. Best bet: driverless out on the ‘open road’ (interstates) where all traffic on your side of the divider is going in the same direction, and an automatic switch to manual mode the moment you cross a city’s outer markers (a sort of reverse ILS), to enter the built environment.
Cheers, Ralph Panhuyzen | @NextGenEV
