Driverless cars need brain science
since brain evolution WORKED hundreds of millions of years ago
Last time I wrote about how robotics needs to get to the level of animals because animals already have 90% of the functionality needed for autonomous cars. By using Patom theory, a model of brains based on neuroscience, new approaches to engineering can be tested.
Applying robotics technologies to the problems of autonomous cars would address the long tail of (insurmountable) problems.
Today I want to review the proposed levels of autonomous cars and compare them to a realistic model of the levels of robotics.
6 levels of autonomy
The Society of Automotive Engineers (SAE) has defined 6 levels of autonomous driving as follows:
Level 0: no driving automation
Level 1: driver assistance
Level 2: partial driving automation
Level 3: conditional driving automation
Level 4: high driving automation
Level 5: full driving automation
There is a description and chart on the FAIST site here (click).
It’s nice to see the numbers aligned with a computer scientist’s model - starting counting at 0!
Level 0 includes technologies that aren’t autonomous as they only support the driver — like ABS and cruise control.
At the other end of the scale, level 5 cars won’t have steering wheels or pedals so the passengers are just cargo, free to enjoy the trip.
As you will see, these 6 levels are for marketing purposes since most don’t relate to autonomy, but create a false appearance of rapid progress. If the first 4 levels are not autonomous because a driver drives the vehicle, why are they in the list at all?
Scale 0-5 is not linear (regarding autonomy)
The scale is a good sales tool. 0 isn’t autonomous at all, and there is an increase in sensory function and motor control as the numbers increase.
Let’s consider the levels from the perspective of how much sensory and motor control is needed for each level. Let’s call this AI for the rest of this newsletter if the car is not driven by a human.
The first 4 levels have a human in control, so they are not autonomous. Therefore features 0-3 need no AI. Level 3, the 4th level, has a driver who can turn over control to the car such as in a traffic jam. The automation can assign control to the human at any time. Therefore, the AI skills needed are limited by the situation. If the car is moving slowly due to traffic and the distance between the cars can be measured with sensory tools like radar/lidar no AI is needed (0%). If the car remains under human control, no AI is needed (0%).
The last 2 levels, 4 and 5, can operate in cars with no steering wheels or pedals. In that case, the vehicle needs AI (100%). So the levels of autonomous cars amount to no AI for 0-3 and full AI for 4 and 5.
Using AI
At levels 4 and 5, AI is needed where the vehicle has sensory apparatus and motor control without human intervention. The same types of skills for safe navigation around the house is needed, with the recognition of situations and objects necessary. If AI cannot recognize animals, humans, children, signs and so on instantly, it isn’t ready to move in the real world with human-like safety.
An autonomous car can therefore be defined as needing AI, with functions in human-like timeframes or accuracies that are not currently technically possible.
The solution for safety at human levels can be approached as proposed by the well-known computer scientist Andrew Ng who reportedly said to remove all humans from other vehicles for safety. Or alternatively we need to build AI that works in less dangerous situations like in the family home and ten when that is working, restart the focus on autonomous vehicles.
Conclusion
Autonomous cars need AI that includes animal-level skills like the recognition of objects and their category (cars, cats, babies, walls, …) and motor control that interacts correctly (don’t drive into walls, wet cement, signs, airplaces, …)
The levels of autonomous cars today are set to either need AI or not. Without AI, a human is driving so these tools are augmentation of human driving only. But when AI is needed (the assumed definition of a driverless car) the long-tail of expertise is needed that isn’t yet available.
A milestone for the safety of driverless cars will occur when the simpler task — human-like robots successfully working in a house — is reached. The industry should work on simpler problems that support the more difficult problems.
Do you want to read more?
If you want to read about the application of brain science to the problems of AI, you can read my latest book, “How to Solve AI with Our Brain: The Final Frontier in Science” to explain the facets of brain science we can apply and why the best analogy today is the brain as a pattern-matcher. The book links in the US are here on Amazon.
