Elon Musk has said that buying a car today that will not be upgradable to enable full autonomy is like buying a horse. It’s implied that this means you should buy a Tesla, since we’re not aware of any other cars on the market that will be upgradeable to full self-driving capability. But one topic that doesn’t get discussed much is electric versus gasoline self-driving cars. Will people be able to buy gasoline self-driving cars? Well, I imagine Elon would call a self-driving gas car a horse with a carrot on a stick in front of it. Let’s explore why.
This article is a non-technical spinoff article from the Tesla self-driving computer analysis I recently published. Elon Musk has said that Full Self-Driving could only work in an electric car. Even today, that sentence really confuses some people because it hasn’t been explained in much depth. The general point is that developing self-driving technology for a gas car would be a huge waste of resources because it is a financial and practical dead end, and there are a couple of major reasons that’s the case other than the obvious “electric car are the future” one.
Gas Cars are Like Electric Cars with a Lot of Lag (Gamer Term for Latency Problems)
A computer can react way quicker than a person can. There are only two things that a gas car can do quickly — slam on the brakes and turn the wheel. Humans have to adapt to this by adjusting their plans to the latency of the car’s gas pedal. They have to get used to the notion that their action won’t result in an immediate reaction. With a mechanical automated transmission a lot of things need to happen: Mechanical torque vectoring with the differential gearbox; altering rpm of the engine every time another transmission is selected; regulating rpm by pumping more or less fuel into the engine. None of that is instant. As I said, it’s like gaming with lag.
It doesn’t matter how awesome a self-driving computer could potentially be if there is a big disconnect between its command, the action, observation & assessment of the result, decision making for the next action, and implementing the subsequent action. The gas car is the weakest link in the chain — or, more accurately, it is the bottleneck when it comes to responsiveness. With electrically controlled torque vectoring, an electric powertrain can react faster than a human driver could, which leads me to the next point.
Safety on Slippery and Icy Roads
This is another topic Elon has touched on before but is a rather complicated topic to explain, and there are no simple, direct, absolute numerical metrics (at least, not published for the public) to help compare the difference in safety between a gas car and an electric car on slippery and icy roads.
First of all, it’s important to understand that, while the rotation of a car’s wheel can be considered an output, a reaction, it is also a sensor that can tell how fast each wheel is spinning and how much traction you have on the road. Now, what can you do with this data? When it comes to a gas car, the only thing it can do is adjust general traction control settings for slippery road conditions, notify the driver of the change in circumstances, and hope for the best, because by the time the gas car senses anything, it’s too late to respond well because the time it takes for the car to respond would be too long to be of much use. On the other hand, besides alert the driver and tune the general settings, the moment a wheel is about to lose traction, an electric car can immediately change the speed at which that wheel is spinning or adjust the other 3 to compensate. The time between observation and reaction is short enough to actually make a difference. (This is not to be confused with a system like “Active Braking System” (ABS) that tries to stop a car when it has already clearly lost control for more than a second.)
What we know for sure, despite lack of data, is that having an electric car makes a difference. The only question is the magnitude of the difference? Does it mean that an electric rear-wheel-drive car is as good as a non-electric front-wheel-drive car in slippery conditions? Or a non-electric all-wheel-drive car? Aside from explaining the theoretical reasons why electric cars are much better in this area, all that we can say for sure is that people have felt the difference and feel more confident about driving a Tesla in the winter, even a rear-wheel-drive Tesla.
Elon’s Remarks about Horses and Why the Safety Benefit is Simple Chess
As was mentioned in the beginning of this piece, Elon has said that if you buy a car that isn’t electric and is not upgradeable to self-driving, then you are practically buying a horse. It has a limited move-set in its arsenal when it comes to dodging accidents. In chess, it’s closer to the move-set of a horse, while a Tesla’s move-set is closer to a queen.
An electric car has instant torque, which means that if a car has to take drastic evasive action to avoid a crash, it can also consider options that require it to instantly accelerate, something a gas car can’t do. This could also apply to swerving out of the way when traffic in parallel lanes might require instant acceleration to successfully merge into another lane, which may be the only option to avoid the crash, and may only be an option if the car can do it very quickly. In such a case, your carrot-on-a-stick-led horse cannot prevent itself from being rump-ended.
Aerodynamic Drag, Motor Efficiency, & Power Efficiency
Compared to gas cars, range is a bit of an Achilles heel for electric cars (for now). The gap is closing really quickly, but in any case, efficiency is a big key to a longer range. This means efficient motors, good aerodynamics, and not wasting too much electricity on other functions, like very-power-hungry self-driving computer technology.
In March 2018, it was announced that the Jaguar I-PACE would be joining Waymo’s fleet of autonomous vehicles. However, more than a year later, not a single Waymo I-PACE vehicle has started commercial operation, and there is probably a very good reason for that.
The Jaguar I-PACE’s real-world range seems to be just a bit below 200 miles (320 km), even though it has a battery larger than a standard Model X vehicle, which has 255 miles of range. Now, add to that the aerodynamic drag that Waymo’s equipment adds to the equation and the processing power needed to make its system work. For a Tesla with computer hardware version 3 (HW3), the power requirement is approximately 100 watts, but Waymo’s equipment might use much more. During Tesla’s Autonomy Day presentation, it was said that in non-highway circumstances the Autopilot computer can account for 20% of used power. If this is also the case for the I-PACE, then its range will be 160 miles (260 km) — but it will actually be worse since that 20% doesn’t include the additional aerodynamic drag from Waymo’s contraption. This brings us to the next point.
How Much are Others Prioritizing Energy Efficiency?
It seems that Tesla is the only company designing self-driving systems very directly for specific models. Other companies and teams appear to be more disjointed — mostly because the primary work has been done or is being done by non-automotive startups, rather than at automobile companies with the automobile design and development teams working very closely with the self-driving technology team. How much the self-driving tech teams have prioritized efficiency of their systems or understood how that would differ in different vehicle models is an open question.
As was extensively covered in our HW3 chip deep dive, a processor needs to be very powerful, but also very energy efficient. In the case of HW3, that means drawing 100 watts or up to a whopping 20% of the power utilization of a vehicle. In the case of Waymo, we don’t know how much power its computer(s) consume, so it could actually be beating Tesla, but since Waymo hasn’t shared any hardware details, there is no way to know.
One thing is for sure — the products that NVIDIA makes are extremely power hungry. Its most current hardware consumes 500 watts for 320 TOPS (which, if we understood NVIDIA correctly from its blog post, can be scaled down to 250 watts for 160 TOPS). Tesla can achieve 144 TOPS with 100 Watts. Basically, where NVIDIA delivers 0.64 TOPS per watt, Tesla delivers 1.66 TOPS per watt. Just for fun, let’s remember that 100 watts in some situations can account for 20% of the power utilized in the vehicle. If that was 500 watts, then FSD would be nearly doubling the power usage of a vehicle (versus no FSD tech at all). Now, I must give NVIDIA some credit — its current product line is a general, multi-purpose product more designed for developing and testing the product, and NVIDIA is promising that a much better chip is right around the corner.
Nonetheless, the main point is that when it comes to self-driving technology, power efficiency might be the second most important metric after safety, and we have no insight into how much consideration it is given by other automakers or chipmakers.
Electric vehicles can respond much more quickly than gasoline vehicles in various ways, which makes them better suited for full self-driving tech. They can take advantage of the rapid response time of computers. The exact optimal combination of steering, braking, power to each wheel, and suspension coordinated by the computer is worlds apart from what is possible with a mechanical internal combustion engine powertrain.
Precision control over the electric motors also allows every single wheel to adjust to slippery road conditions, making a rear-wheel-drive vehicle almost as safe (or maybe even safer) on icy road conditions than a gas car with front-wheel drive or all-wheel drive.
The combination of a computer AI driver with an electric car is so much more versatile and can react so much faster to traffic situations that it just does not make sense to continue producing (or buying) gas cars.
However, it is critical to design efficient self-driving tech and efficient vehicles, as well as designing the two to integrate as efficiently as possible.
That was fun. I hope you agree. Let’s not even open the Pandora’s box of possibilities the Roadster 2 opens by possibly being able to turbo boost a few meters into the air to avoid an accident. 😉