As noted in my previous post, I’ve been doing work on the role of A.I. in design, since my work on thermodynamics lends itself naturally to the topic. In particular, I’ve been considering how A.I. could be used to improve automobile design, and I’ve come up with several ideas that I think could prove both useful, and quite fun.
Dynamic Engine Sizing
The first idea is to use a dynamic engine size, that is adjusted by the height of a fluid, that raises and lowers the floor of the combustion chambers. This would allow the pressure inside the chamber to be adjusted dynamically. So for example, in an environment where you don’t need a lot of power, you can raise the floor of the chamber, thereby shrinking the size of the chamber, which will achieve higher pressure, without much gas per cycle. In contrast, if you’re in an environment where you do need a lot of power, you can drop the floor of the chamber, allowing for more gas to be injected per cycle.
The big picture idea would be to have all of the controls within the engine, including the chamber size, monitored and controlled by a machine learning algorithm, that can then optimize the performance of the engine, given the context in which it is operating. This would allow for different modes of operating, each informed by datasets. For example, if you want smooth acceleration, you want to minimize jerk (i.e., the derivative of acceleration), which is something that can be tested for prior to the release of the vehicle. Similarly, you could imagine another mode where you instead maximize power, whether for recreation, or because you’re pulling a hitch, or climbing a hill. Or, you could instead simply maximize fuel efficiency. The amount of data available to inform how the engine behaves is likely to be robust, since it’s not difficult or expensive for a vehicle to determine where it is, what the angle of ascent is, what the weight on the frame is, etc., allowing for meaningful comparisons to prior data.
As a general matter, using machine learning to manage the fuel injectors, spark plugs, engine pressure, etc., would allow vehicles to be released with useful datasets produced by testing, which can then be updated based upon new observations derived from the owner of the vehicle actually driving, while the car is not in use.
The mass of a car is an important variable to control for, both for fuel efficiency, and safety. Specifically, the more massive a car is, as a general matter, the less fuel efficient it is, because you need more energy to move a larger mass. In the context of safety, the more massive a car is, the more resistant it will be to acceleration, which implies that it’s less likely to roll in an accident, which is, as a general matter, a good fact for the safety of the passengers.
What I realized is, if the inertia of a car comes from a set of gyroscopic stabilizers, then you can dynamically adjust the inertia of the car. This would allow you to start with a very lightweight car, which is more fuel efficient, and then add inertia dynamically dependent upon context. So for example, if you’re on the highway, traveling at a high speed, and you’re in a lightweight car, then you probably want additional inertia, which will prevent you from rolling in an accident. Again, these stabilizers could be controlled by machine learning algorithms, that would adjust them depending upon context. Though I can’t be certain it would work, you could even imagine shifts in inertia, in combination with other conditions also controlled by A.I., being used to break a hydroplane, which could also improve safety outcomes. The point being that if a person thinks they’re in a truly hopeless situation, they can hit a button, and the machine takes over, and attempts to restore the initial conditions. You could even imagine these circumstances using datasets trained by expert drivers, so that, e.g., if you hit a hydroplane, when you hit the button, the dataset drawn upon uses examples of professional drivers managing similar circumstances.
This would also allow you to adjust the distribution of inertia throughout the vehicle. So for example, if you have a very light frame, with an extremely heavy engine, which would be ideal for a sports car, you could redistribute the inertia using gyroscopic stabilizers, by adding inertia to other areas in the vehicle, improving stability. Or, if you have an extremely heavy item in your trunk, then you could add inertia to the front of the car, which could improve stability.
Dynamic Environmental Controls
Basically every modern car has environmental controls that allow you to adjust the temperature of the cabin using heating and air conditioning. However, they’re typically set to a level, rather than a temperature. For example, the air conditioner in a car generally runs from low to high, rather than being expressed as a temperature. If however, you include a series of thermometers inside and outside the cabin, you could instead let the passengers express the cabin conditions as a temperature, and then let a machine learning algorithm control the heating and cooling systems, with the goal of achieving a particular temperature outcome, based upon the data. This would allow the passengers to simply state their preferences, and allow the algorithms to achieve them, and moreover, achieve them in an efficient manner, based upon data. As a general matter, the big picture idea is for a passenger to express a goal state condition (i.e., a particular temperature, or level of internal or external luminosity), and then have a machine determine the appropriate settings necessary to achieve that desired goal state condition.
As a design matter, personally, I think the vents in a car are disgusting, both aesthetically, and as an experience, with a bunch of hot or cold air from the depths of a car blowing in your face. I think it makes far more sense to simply perforate the cabin of the car, and have temperature-adjusted air circulated through those perforations, which can be placed in less invasive locations.