Running On Empty Could Befall Self-Driving Cars

Dr. Lance Eliot, AI Insider

Image for post
Image for post

[Ed. Note: For reader’s interested in Dr. Eliot’s ongoing business analyses about the advent of self-driving cars, see his online Forbes column: https://forbes.com/sites/lanceeliot/]

Running out of fuel.

It’s a pain in the neck.

During my college days, a friend of mine had an old jalopy of a car that was busted-up and yet it still managed to passably work, but the gas gauge seemed to have a mind of its own.

I say this because sometimes the fuel gauge would be showing that the tank was half full, even though it was nearly empty, and at other times the gauge needle was on empty but there was actually a half of tank or more of gas in it. It was a wild guessing game as to how much fuel there really was in his car at any point in time.

Some affectionately referred to it as the Guess-o-Meter.

Many people don’t realize that allowing your car to run out of gas can be bad in various ways, beyond just getting stranded someplace.

The car doesn’t especially like the notion of you letting the gas entirely be used up. The gas lines can get air in them, which can make it hard to restart the car once you’ve gotten more gas into the tank. Other things can go wrong with the engine by running out of gas.

CARS RUNNING OUT OF FUEL

According to statistics reported by the American Automotive Association (AAA), they respond to about 16 million or more calls per year for being out of fuel.

Of course, that number is only accounting for those that have the AAA service.

How many people per year really get stranded by running out of fuel?

I’d bet it is an even bigger number.

When I refer to running out of gas, I am also alluding to running out of electrical charge too.

In other words, running out of fuel encompasses both gasoline powered cars (often referred to as ICE or Internal Combustion Engines), and also electrically charged cars.

Colleagues that have Electrical Vehicles (EV) and sometimes say they are low on gas, even though they know and I know that it is not gasoline but instead an electrical charge.

One of the drawbacks right now with EV’s is the aspect of finding a place to charge your EV.

Given the somewhat narrow range of miles that today’s EV’s can go on a single charge, you need to be mindful of where the charging stations are. Unlike gas stations that seem to sit on every corner, you aren’t going to as likely be able to find a place to charge your car.

This is gradually changing as more charger locations are established, but for now, it can be a bit dicey to be using an EV for any kind of long-range driving.

Some of the tow services are now carrying with them a fast-charger to aid EV’s that have run out of fuel.

RUNNING OUT OF FUEL WHILE UNDERWAY

One of the extremely dangerous aspects about running out of fuel involves a car that is in-motion, which presumably is mainly when you would discover you are out of fuel.

Without fuel to run the engine, your car now becomes a danger to those in it and those nearby to your car. You might be able to coast for a little while, but your overall ability to maneuver is now dramatically stinted.

Don’t you know how dangerous your coasting car can be?

What kind of a person let’s this happen?

Tow truck drivers often get the wildest stories from those that have gotten stranded and are out of fuel.

Some culprits will just admit they ignored the warnings.

In any case, we have the grave danger of a car that has run out of fuel and that is in the midst of being on our roadways and has the potential for becoming an unguided missile.

I say unguided and realize you might object and say that the coasting car should be able to be guided by the driver.

Keep in mind that for some cars, trying to “guide” a moving car that has run out of fuel is not so easy. Also, you no longer can presumably use acceleration to get yourself out of a pickle.

AI AUTONOMOUS CARS AND RUNNING OUT OF FUEL

What does this have to do with AI self-driving driverless autonomous cars?

At the Cybernetic AI Self-Driving Car Institute, we are developing AI software for self-driving cars. One aspect of interest is what to do about getting low on fuel and how the AI should deal with such a matter.

Allow me to elaborate.

Presumably, the AI will be able to detect the amount of fuel available in the self-driving car.

I’ll for the moment put to the side the circumstances wherein the fuel detection system is faulty.

This could happen and I just want to point out that it could happen. This is sometimes a surprising notion to those that believe that AI self-driving cars will somehow have super powers and never fail.

Hopefully, the AI should be developed such that if the self-driving runs out of fuel, even if the fuel detection system is claiming there is fuel, the AI will have a fallback operation that can then try to get the self-driving car into a minimal risk condition, such as getting over to the side of the road. I realize that the odds of the fuel detection system being so far off that the AI gets caught unawares would seem generally unlikely, but the fact is that it could happen and thus there needs to be a contingency by the AI for this possibility.

We can also ponder whether the AI would even be able to realize that the fuel is exhausted.

Imagine that the fuel detection system is informing the AI that there is a half tank full of fuel. Meanwhile, the self-driving car starts to sputter and the engine dies. The AI now has a situation of the self-driving car’s engine no longer working, but the AI is getting meanwhile no particular indication of why. The fuel volume seems to be just fine, and yet the engine has shut-off.

There is another possibility too involving some other mechanism that has gone awry and won’t feed fuel to the engine.

Thus, the fuel detection system might be completely correct, and yet the fuel getting from the tank or storage bank to the engine has gone afoul. The point being that there are more than just one means by which the engine might no longer be getting fuel, either due to the lack of fuel or the inability for the fuel to actually reach the engine.

The AI self-driving car should have various sensors and electronic communication going on with the ECU (Engine Control Unit), which then might aid the AI in diagnosing what is going on. This though can be tricky since presumably it has to happen in real-time. Suppose the AI is driving the car on a freeway at 80 miles per hour and all of a sudden, the engine stops running.

There is a need for the AI to try to quickly figure out what has gone amiss. At the same time, no matter what has happened, the AI needs to be taking action about the aspect that the engine is no longer working. You can debate somewhat about how much diagnosis is needed other than knowing that the engine has stopped. On the other hand, if the AI could discern why the engine has stopped, it might open more possibilities of what next action to take.

COPING WITH FUEL ISSUES

In short, we assert that the AI needs to have a capability to contend with a situation involving the AI self-driving car running out of fuel.

Some AI developers would say that this is covered by their general approach of having the AI deal with anything that might go amiss with the AI self-driving car. In essence, they claim that running out of fuel is no different from say the self-driving car having a tie rod that breaks. It’s all part of the contingency driving aspects routine.

This is not really the case.

If the car is otherwise fully functional, other than the lack of fuel, there is a chance to use the in-motion of the self-driving car to try to cope with the situation. A severe mechanical failure is unlikely to allow the AI the chance of maneuvering the self-driving car and trying to find a safe way to get out of the current difficulty. We eschew this notion that all issues are the same and that there is no need to differentiate the various kind of issues that can arise during a driving journey and the functional capabilities of the self-driving car.

Let’s shift gears, so to speak, and now consider the situation of an AI self-driving car that is genuinely getting low on fuel and the AI knows that the self-driving car is indeed getting low on fuel.

What happens then?

Some auto makers and tech firms are skirting the issue right now because they are essentially controlling their AI self-driving cars, and so they force the AI to go ahead and do a fill-up.

In other words, if you are doing roadway trials with your AI self-driving cars, and you are pampering those AI self-driving cars with a dedicated team of engineers and maintenance personnel, the odds are that you aren’t going to allow the AI self-driving car to run out of fuel.

In the wild, once AI self-driving cars are prevalent, and in the hands of consumers, what happens then?

We’ll consider two different circumstances, one involving human occupants inside the AI self-driving car when the fuel question arises, and the other is the situation when there is no human inside the AI self-driving car. Keep in mind that AI self-driving cars will be driving around at times with no human occupants. It could be that the AI self-driving car is trying to get to location where it is going to pick-up humans, or maybe it is being used as a delivery vehicle and so heading to a destination to make the delivery, and so on.

It is predicted that AI self-driving cars will be used extensively for ride sharing purposes. If you owned an AI self-driving car that was being used for ride sharing, you might have it trolling around as it waits for a potential rider to request a ride. You look out the window of your office and see your AI self-driving car cruising back-and-forth, waiting for someone to request a lift.

Why not park the car?

There might not be any available parking, plus you might want your AI self-driving car to be the first to respond to a request, and if it is cruising around this might be a better chance than if it is stopped and parked.

PASSENGER INTERACTION

If an AI self-driving car has human occupants, and the AI detects that the fuel is getting low, should the AI let the passengers know?

If so, should it ask them if it is OK for the AI to then find a place to get more fuel?

It would be as though you are in an Uber or Lyft ride sharing car of today, and the human driver turned to look at you, while sitting in the backseat as a passenger, and let you know that the car is getting low on fuel. This is informative to you. The driver might then ask if it is OK for the driver to stop at a nearby gas station and fill-up. I’m betting you would be irked by such a question. You are presumably paying to get from expeditiously from point A to point B. Having to stop and have the car get fuel seems like a rather untoward act.

Suppose you refuse the request by the driver.

If the knows they cannot get to your destination prior to running out of fuel, it would likely that the human driver would insist that the car must be brought to a fueling station and whether you like it or not, the driver is going to do so.

Outrageous, you exclaim!

Why didn’t the driver beforehand make sure there was sufficient fuel to make it to the desired destination?

This is the same logic that some AI developers tell me about the question of fuel when I ask them what they are doing about fuel levels in their AI self-driving car software. These AI developers tell me that it won’t ever happen that the AI self-driving car will have passengers and be at a low ebb in terms of fuel. The AI is programmed to always be getting more fuel whenever it is otherwise not engaged with a passenger, plus, the moment that a passenger indicates where they want to go, the AI can ascertain whether there is enough fuel and thus refuse to take on the passenger if the fuel would be insufficient.

There are some holes in this logic.

Suppose that an AI self-driving car has taken on-board a passenger that wanted at first to go to the nearby park. The AI calculated the miles involved and figured out how much fuel there is and determines that it can make it to the park. During the driving journey, the passenger says that they need to pick-up their friend that also wants to go to the park, which means a side trip now for the AI self-driving car. Imagine that this then pushes the potential fuel consumption such that the AI self-driving car would not be able to get to the friend and to the park and then to a fueling station.

You’ve now got a circumstance of a passenger in the AI self-driving car and there is insufficient projected fuel to satisfy the driving journey being requested. Thus, this fanciful notion that the AI self-driving car would never have a passenger and yet get into a situation of not having enough fuel for a driving journey is shall we say weak.

My point is that there is going to be situations in which the AI self-driving car will have passengers and yet the desire of the passengers might exceed the projected available fuel.

The AI should be able to then interact with the passengers and explain the situation. Furthermore, there might need to be an interactive dialogue about what to do. In the case of the side trip to get the friend while on the way to the park, perhaps the AI explains that it needs first to fuel-up if the side trip is to be undertaken, and therefore “negotiates” with the passenger about what to do.

You might say that it is unnecessary to have a dialogue with the passengers and instead the AI can just tell the passengers what is going to happen, regardless of actually what they want or doing any kind of interaction with them. The AI might simply tell the passenger that wants to go pick-up a friend prior to going to the park that this side trip is not going to happen. Without even particularly saying why, the AI might just emit an indication that the side trip is not being permitted and that’s that.

I have a feeling that if AI self-driving cars do that kind of dictatorial driving, people are not going to want to get into an AI self-driving car. People would probably be more willing to deal with a human driver than to have a “robot” that tells them what is going to happen and offers no capacity to try to explain or reason about the driving.

On the other side of this coin, presumably we don’t want a human passenger to let the AI self-driving car get into a dicey situation.

Suppose the AI tells the passenger that the side trip will mean that the AI self-driving car would run out of fuel and get stranded. The passenger maybe decides to tell the AI to go ahead and proceed anyway. Is the passenger crazy? We don’t know. Maybe the passenger misunderstood the AI and the situation. Maybe it is an emergency of some kind and the risk is worth it to the passenger? Could be various explanations.

Anyway, should the AI allow a passenger, a human, the ability to override what the AI has determined to be the case that the AI self-driving car is likely going to run out of fuel and become stranded, and thus the human is telling the AI that it is Okay to have this happen?

If you are the auto maker or tech firm, you likely would say no, never allow a passenger to override this kind of circumstance. If you are the human passenger and have some kind of rationale for why you want to do this, you are going to perhaps have a different viewpoint on the AI essentially overriding your command to it.

I’ll point out that this is not the only boundary of having the AI and a human at potential odds about a driving task.

There are other kinds of situations in which the AI is going to want to do one thing, and a human passenger might want to do something else. The industry does not yet have any clear cut means of trying to ascertain when the AI should so proceed versus acquiesce to the wishes of the human. It’s a difficult matter, for sure.

WHEN NO PASSENGERS ARE INVOLVED

Let’s next consider the situation when there isn’t a passenger in the AI self-driving car.

This would seem generally to be an easier circumstance to deal with in terms of the fuel situation.

Suppose the AI self-driving car is trying to deliver a package across town and at first calculated it could make it to the destination without having to re-fuel. Turns out that the AI self-driving car got stuck in traffic due to a car crash that had blocked all lanes, and the fuel of the AI self-driving car got excessively used up due to this unpredictable and unforeseen delay.

I think we might agree that the AI self-driving car should as soon as practical go get fueled-up.

This is likely to create an even further delay in delivering the package. Presumably, the AI self-driving car is going to be interacting with some other system or people to let them know about the delay.

Overall, whenever there is not a passenger in the AI self-driving car, I would guess that we would expect the AI to be monitoring the fuel and take care to make sure to go do a re-fuel when needed. There would need to be a sufficient safety margin that whatever kind of unexpected aspect arises, it hopefully can still make it to the refueling.

Suppose though that the AI self-driving car is not able to do so, such as the case of being on the freeway and all lanes of traffic are blocked.

In that case, it could be that the AI has no options to do anything other than sit there on the freeway and use up fuel. Sure, it can try to minimize the amount being consumed, but let’s assume that it ultimately does run out of fuel and has no other means to do anything (it is stuck in traffic and no way to get out).

I know some AI developers that claim this kind of scenario is less odds than a meteor flying down to earth and striking the self-driving car. I’m not sure they are right about that aspect. My view is that betting on something never happening and yet that it could happen, seems like a lousy bet.

In an era of AI self-driving cars that are prevalent, presumably the AI self-driving car could use V2V (vehicle to vehicle) electronic communications to let other nearby AI self-driving cars know that it is getting low on fuel. This might then get the other AI self-driving cars to help out and open a path for the AI self-driving car to get off the road.

I suppose that if the AI self-driving car did run out of fuel on the roadway, perhaps the other nearby AI self-driving cars might come to its aid. For example, another AI self-driving car might give the stranded one a push to help it get out of traffic. Other AI self-driving cars might block traffic to let this activity take place.

Meanwhile, via perhaps V2I (vehicle to infrastructure), an electronic message goes out to a local tow truck that it should come and get the AI self-driving car.

CONCLUSION

The Kepler space telescope recently ran out of fuel, bringing to an end its nearly 10-year planet hunting voyage across outer space, having successfully discovered several thousands of exoplanets.

Away it now drifts, looping around the sun and taking on an Earth-trailing orbit.

NASA knew that the Kepler would eventually run dry.

Closer here to home, we should be thinking about the situations when an AI self-driving car goes dry, i.e., running out of fuel. It seems like perhaps a trivial matter in comparison to hunting for new planets, but I assure you that once we have a prevalence of AI self-driving cars, people are going to want to know that those AI self-driving cars can deal with being low on fuel, along with being able to appropriately handle situations of running out of fuel.

I know it seems counter-intuitive that these seemingly “super powered” AI self-driving cars could somehow find themselves stranded, and we would assume that running out of fuel could only happen to distracted or imprudent human drivers, but these kinds of assumptions need to be revisited.

Maybe we’ll see an AI self-driving car that one day is pleading for fuel, hey buddy, the AI asks another nearby self-driving car, can you spare a gallon or two (or, perhaps some megawatts)?

For free podcast of this story, visit: http://ai-selfdriving-cars.libsyn.com/website

The podcasts are also available on Spotify, iTunes, iHeartRadio, etc.

More info about AI self-driving cars, see: www.ai-selfdriving-cars.guru

To follow Lance Eliot on Twitter: https://twitter.com/@LanceEliot

For his Forbes.com blog, see: https://forbes.com/sites/lanceeliot/

For his AI Trends blog, see: www.aitrends.com/ai-insider/

For his Medium blog, see: https://medium.com/@lance.eliot

For Dr. Eliot’s books, see: https://www.amazon.com/author/lanceeliot

Copyright © 2019 Dr. Lance B. Eliot

Written by

Dr. Lance B. Eliot is a renowned global expert on AI, Stanford Fellow at Stanford University, was a professor at USC, headed an AI Lab, top exec at a major VC.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store