Self-Driving Car Difficulties When Driving At Night

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/]

Are you a nighttime driver?

According to various governmental reports, it is said that there is about 60% less traffic at night but 40% of all fatal car accidents happen at nighttime, furthermore the fatality rate for car occupants is about three times higher at night than during the daytime, and the fatal crash rate for 16-year-old drivers is twice what it is at night versus during the day (a good reason to keep your teenager from driving at night!).

What makes nighttime driving so terrible?

One aspect is the impact on visibility.

I suppose if it was only the lighting that made a difference, perhaps we could all accommodate it as a nighttime issue.

But, as you know, there is also driver behavior that comes to play.

Drunk driving is much more likely at nighttime than daytime. This means that a driver is likely not able to properly navigate the roads and can either get into an accident or cause someone else to get into an accident. There are also drowsy drivers at nighttime. These drivers are maybe worn out from a long day’s work, or maybe they need to get up at 3 a.m. to drive to work in the early morning darkness. Being drowsy causes them to be less aware of the traffic, more likely to swerve or take adverse actions, and be more prone to either getting into an accident or causing one.

Additional Difficulties Of Nighttime Driving

I’ll punch things up with even more factors about nighttime driving.

Since there tends to be less overall traffic at night, it also tends to allow drivers to go faster and perhaps cut corners more so.

In addition, I know some drivers that believe they are less likely to get a traffic ticket at nighttime, apparently they believe they are less likely to be caught under the cloak of darkness, so they tell me that they aren’t as worried about being full legal in their driving at night.

Kind of makes you scared of driving at nighttime with that kind of logic among some drivers.

Oddly enough, it seems like pedestrians and bike riders also become riskier at night.

AI Autonomous Cars And Nighttime Driving

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 for self-driving cars and undertaking special attention to the nighttime driving capabilities.

There are some automakers and tech firms developing self-driving cars that consider nighttime driving to be nothing more than daytime driving.

Driving is driving, they say.

We respectfully disagree with that somewhat gross oversimplification of the matter.

Our viewpoint is that nighttime driving requires added capabilities.

First, perhaps the most obvious aspect is the detriment to visibility.

The cameras on the self-driving car will likely not be able to catch as sharp of images and the video will also be less illuminated than were it daytime. The headlights of the self-driving car become particularly vital. If the headlights are in poor shape or occluded due to dust or dirt, it could have an even worse impact on the image processing. The cameras at the rear of the self-driving car are without any direct lighting such as headlights, and so those are obviously severely degraded in terms of the images captured.

The AI self-driving car should be checking to make sure that the headlights seem to be working sufficiently to capture apt images, as best possible, which can be tested beforehand in terms of doing testing at the start of a journey.

The image processing of nighttime images is different than the image processing for daytime images.

Images that are well lit can be more readily processed. Images that have dark shadows and other darkened portions is harder to process. This often involves doing various image processing clean-up. It can also change the probabilities associated with object detection. An object that during daytime had say a 100% probability of being a wheelchair at the curb might at nighttime have a 60% chance of being a wheelchair.

There is also typically much more light reflections and distortions.

Street lights and the headlights can bounce off the car ahead of you, perhaps doing so off the chrome bumper that acts like a kind of mirror, and thus other objects either become more well lit or become overly lit. The bouncing light can actually cause a sunburst effect and make the object recognition harder. Again, the image processing software has to be ready for this impact.

Machine Learning Aspects

For machine learning and the use of artificial neural networks, if those neural networks were trained with pictures of those objects during the daytime, it could be that those objects are no longer detected by the neural network. It’s important the training of the neural network include objects as seen in darker conditions. Some prefer to do both daytime image and nighttime image training together, as a smorgasbord, while others believe that the neural network will be better off if one instance is trained to daytime images and the other to nighttime images.

One aspect that visually can become more problematic is the detection and interpretation of street signs.

Sensor Fusion Aspects And More

During sensor fusion, the sensor results are compared and combined in various ways.

If the cameras are not getting good images, it is usually the case that the AI system is devised to give less weight to those sensors.

Thus, if the camera is suggesting that there doesn’t seem to be a pedestrian at the side of the road (perhaps the image is so darkened that there is no ready way to detect a figure standing in the shadows), but the radar is detecting that something might be there, the sensor fusion has to determine what to do.

Should it rely upon the radar and claim that there is a pedestrian there?

During daylight, often times the sensor fusion opts to wait until the cameras are able to agree with the radar results. This though can be dicey depending upon how much time is available before a decision about the matter needs to be made.

Overall, at nighttime, the cameras are likely to be much less reliable and thus the sensor fusion should be designed to cope with this aspect.

It’s almost the same as a human driver.

If a human driver cannot readily see at night, they are at heightened risk of properly performing the driving task. Purists for AI self-driving cars point out that the aspect of multiple sensory devices on an AI self-driving car presumably implies that therefore the AI self-driving car will be better able to do nighttime driving than a human can (most humans don’t have radar or LIDAR built into their bodies).

The virtual world model that is used by the AI must also be updated and be apprised of the nighttime aspects. This can include that the associated probabilities with detected objects are going to be lower than the normal daytime driving, as mentioned about whether a wheelchair is at the curb or not. And, the AI action plans need to take into account the now somewhat degraded virtual world model, along with what kinds of actions can be taken in nighttime situations.

Suppose the AI decides that a hard braking action is needed.

At nighttime, the driver behind you might not be as ready to stop, versus during daytime they might have more readily been aware of you ahead of them. Too, keep in mind that the human drivers on the roads at nighttime are potentially drunken or drowsy, or cannot see the road and traffic as well as they could during daytime. The AI needs to consider how other drivers will be acting and reacting at nighttime. This might also dictate that the AI needs to be going slower or taking other defensive maneuvers that would not necessarily be needed during daylight hours or used as frequently.

Additional Nighttime Driving Concerns

I’m sure that some proponents for AI self-driving cars will instantly say that there shouldn’t be any human drivers on the roads, and that we need to get to the point of having only AI self-driving cars out there. Thus, there would be no need for the AI to worry about human drivers at nighttime. No more drunken human drivers that leave the bar after midnight and then crash into other cars. No more drowsy human drivers that wake-up before sunrise and are so sleepy when they drive that they ram into a pedestrian.

Instead, nighttime driving will be a breeze.

The AI self-driving car would communicate via V2V (vehicle to vehicle communications) with other self-driving cars, and they would all help each other in nighttime driving conditions. The AI self-driving car ahead of you might have figured out that there’s a pedestrian standing at the crosswalk, which perhaps your AI self-driving car cannot quite determine due to the darkness and angle, and shared this aspect with your AI of your self-driving car. Cross sharing would aid the AI’s in doing enhanced nighttime driving on a collective and collaborative basis.

Sure, this might well happen, but it’s going to be far off in the future. There are about 250 million conventional cars in the United States today. Those are not going to magically become AI self-driving cars right away.

There are more twists and turns about nighttime driving for AI self-driving cars.

Imagine that it is nighttime and there are heavy rains coming down.

When you combine the weather with the nighttime aspects, it can be a potent danger for any driver, human or AI. Another twist is roadwork.

Generally, most of the road work that we have done on our infrastructure is done at nighttime, which is done to reduce the adverse impact to daytime traffic. For the AI to contend with roadwork can be difficult, and especially so at nighttime. Trying to figure out that the street is coned off and you can’t make a right turn up ahead, it’s not easy to do at nighttime.

Conclusion

As they say, do not go gentle into that good night, but instead rage, rage against the dying of the light.

We need to have AI systems for self-driving cars that realize there is a difference between daylight driving and nighttime driving.

Treating daylight as the same as nighttime, or assuming that nighttime driving is equivalent to daytime, it’s a dangerous trap that could cause a grave and overbearing darkness to descend upon the future of AI self-driving cars.

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