Driving In Reverse Is A Surprisingly Tough Challenge For Both Human Drivers And 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/]

One of the worst nightmares for any driver is the chance of backing up a car and running over someone.

When you are backing up, it can be very difficult to know what’s behind the vehicle.

There’s a famous video of a crawling baby in Brazil that unbeknownst to the parents crawled behind the family car as it was being backed out of the garage. The car got about halfway over the top of the baby when a person walking nearby pointed out there was a baby underneath.

Getting out of the car, now stopped over the baby, the family members were fortunately able to pull out the baby and did so without much harm having come to the child.

They were lucky.

Statistics were against them in the sense that by-and-large once a backout is underway, whomever is getting hit is likely to be severely injured or even killed.

Per federal data, in the United States alone there are over 200 deaths annually and more than 15,000 persons injured via backover incidents.

As you might guess, young children in the age of 5 or less account for nearly one-third of those deaths (of those, mainly children in the 1–2 years old bracket), while adults over the age of 70 are about one-quarter of the deaths. In essence, very young children and older elders are the most likely major segments of being the victim of a backover death.

Backup Cams: It’s The Law

I’m sure that you are thinking that if we had backup cams on cars then we wouldn’t have any of these deaths and injuries.


First, you might find of keen interest that after years of delays in implementing a law passed by Congress in 2008 requiring regulators to enact legal measure that would require auto makers to enhance rear view visibility, some ten years later, there is indeed a requirement that new cars being sold in the United States must be outfitted with backup cameras (the new requirement was announced in 2014 and the car makers were given four years to implement it, starting in 2018).

For those of you with modern cars, you’ve already likely got a backup cam in your car, so this law doesn’t mean much to you, other than the aspect that gradually there will be a lot of cars with the backup cam.

Eventually, once older cars end-up on the junk heap, all cars will have backup cameras as the newer cars become the dominant proportion of all 200+ million cars of today (this will take many years though to playout).

I guess we can call the “backover” problem solved since we’re going to have all these backup cams — if you believe this you are in for a bit of a surprise.

Turns out that a study in 2016 found that of cars outfitted with a backup cam and the same model of cars without a back-up cam, there was only about a 16% drop in reported backover incidents for those cars with the back-up cam.

Think about that for a moment. You might have assumed that there should be a 100% drop in backover incidents. Having a back-up cam implies no more backovers.

Backing Up Is A Driving Weakness

Well, a backup cam is only as useful as the nature of the driver at the wheel.

It is unlikely that all drivers will actually look at the display in their car to see what the backup cam shows them.

For some people, they get so used to the backup cam that they rarely look at it.

Even if the driver does look at the backup cam, they might not notice what the backup camera display is showing them.

Another factor is whether an object moves into the field of vision at the last moment.

Automation To Aid The Backing Up Driver

If the backup cam alone won’t get the job done, I am sure you are thinking that let’s put some automation onto the task.

Indeed, there are some backup cam systems that have an alerting feature. If the backup cam detects an object in the field of view, it will make a tone or some other alert inside the car to let the driver know. This could definitely help for those drivers that aren’t rigorously always studying their backup cam.

In addition to an alert, or in lieu of an alert, another kind of automation is an emergency braking system for backover prevention or mitigation purposes.

AI Autonomous Cars And Backing Up

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

At the Cybernetic Self-Driving Car Institute, we point out that by-and-large self-driving cars are going to be equipped with sensors that allow for looking behind the car, and so it is an already built-in capability that just needs to be leveraged by the AI of the self-driving car. That’s an “edge” problem that we are working on.

An edge problem is considered one that is not at the core of something.

The core of an AI self-driving car is having the AI be able to drive the car forwards, and be able to drive down streets, drive on freeways, make safe left turns, and otherwise do all the things a human driver can do.

In essence, right now, the auto makers and tech firms are mainly concerned with making an AI self-driving car that can drive forwards. Going in reverse is considered a secondary problem, or what some refer to as an edge problem, since its not at the core of the driving task (as they view it).

The AI self-driving car pretty much should already have the needed sensory devices, and so the other aspect is the AI software to leverage those sensors.

That being said, not all of the emerging AI self-driving cars necessarily have a typical backup cam per se.

They might have other cameras on the rear of the vehicle, though not necessarily of the type for a backup purpose and nor aimed at the ground behind the vehicle.

Some of the cameras are instead aimed at a further distance, so as to detect a car behind the self-driving car.

One added potential plus for an AI self-driving car is that there are usually radar, sonar, and LIDAR on the car too, which can be used in combination with the cameras.

I want to point out that very important element that I just mentioned.

A conventional car that is outfitted with a backup cam is unlikely to have radar, sonar, LIDAR, and other sensors that can be used in combination with the backup cam. A conventional backup cam is all alone. It is the only means to try and detect what’s behind the car. This is slim.

We’re incorporating the other sensors into the gambit of preventing backovers so that we can increase the chances of avoiding a backover, doing so by bringing together the visual data, the radar data, the sonar data, the LIDAR data, and the rest. The AI of the self-driving car has to be doing some solid defensive driving when backing up.

Why Backing Up In Cars Is Vital

It is helpful to consider the major use cases associated with backing up a car.

We back out a car in relatively common circumstances.

There are exceptions beyond the common circumstances, but it’s best to focus initially on the common ones and then branch out from there.

First, there is backing out of a garage as a driving-in-reverse type of task.

This is extremely common for people to want to do.

Next, there’s the act of driving down a driveway while backing out.

A similar use case is driving up a driveway while backing out.

There’s backing into a parking spot as another commonly performed task, and likewise backing out of a parking spot (less frequent, but sometimes combined with going back-and-forth to inch out of a tight parking spot).

Throughout any of those backing up operations, the AI needs to be on the watch for obstructions.

The obstructions might be large or small. They might be still or in motion.

There are also some important exceptions.

For example, suppose there is a person purposely standing at the back of my AI self-driving car that is warning other people to stay clear.

This helpful person, they themselves now become a detected obstruction by the AI. How will the AI know that the obstruction is actually part of the backing up operation?

Exceptions To Backing Up That Allow Driving Over Something

Each morning, I dutifully back down my driveway and drive over the newspapers that are laying on my driveway, having been earlier tossed there by the newspaper delivery service.

When I get home at night, I once again drive over the newspaper, and upon parking my car in the garage, I get out of my car and go to get the newspaper on the driveway.

Let’s for the moment assume that the self-driving car sensors and the AI are good enough to be able to detect that the newspaper is sitting there on the driveway.

The AI would presumably refuse each morning to backup and later when I get home would refuse to go forward, since in both cases I am driving over something.

Thus, this is a harder problem than it might seem, since there are going to be circumstances where the human occupants want the AI self-driving car to proceed with backing up, in spite of a potential rollover of something or the nearness of a human or other object.

The human occupant will need to have some means to communicate with the AI self-driving car, such as by using an in-car command. This human directive capability though has a downside, since suppose the human occupant intentionally wants to harm someone and so tells the AI self-driving car to proceed to backup into the person — should the AI self-driving car comply? As you can see, there are ethics issues involved in this too.


We know that backup cams are coming to conventional cars, slowly, gradually, and that some cars will also have alerts or emergency braking systems combined with a backup cam.

Older cars are unlikely to have this.

Only some of the newer cars will have it.

For AI self-driving cars, they are destined from the start to have sensory devices that can be leveraged for backing up safely.

We just need to make sure that right kind of sensors are being included, and that the AI is savvy enough to leverage those sensors.

I’d like to be able to say that with AI self-driving cars we’ll have eliminated the backover problem, but realistically it won’t eliminate it, but at least it should help to reduce the frequency and magnitude of backover incidents.

Let’s not back out of that goal.

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 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