Self-Driving Cars Can Be Dangerous For Bicyclists Along With Being A Savior
Dr. Lance Eliot, AI Insider
[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 living in Biketown or in Bikelash?
Let’s start with Biketown, which is any locale that welcomes bicycling.
Bicyclists, some would say, are wonderful because they are green, meaning they are good for society by using a non-polluting form of transportation.
Many cities have opted to increase the number of bike lanes that they provide.
Some cities even have specially painted traditional car lanes to indicate that those lanes are intended for bicyclists to ride in.
It’s considered the “last mile” of ridesharing (you use a car-based rideshare to get near to a desired location, and then bike the remainder of the way rather than walking).
It’s not all roses though in the biking world.
Let’s consider Bikelash, consisting of those that have serious qualms about bicyclists and bike riding.
According to published statistics, there are an estimated 45,000 bicyclists injured each year in reported roadway accidents (that’s the reported number, while the true full number including unreported incidents is likely much higher).
The number of bicyclists deaths seems to range anywhere from 800 to 1,000 per year, and some numbers suggest that it really is more like 3,000–4,000 if you also include severe injuries that leave the bicyclist maimed for life.
In short, anytime you get onto a bike, you’ve just increased your odds of injury or possibly death. Don’t want to be sour on bike riding, and I’m just trying to emphasize that it’s a dangerous “sport” and we often take it for granted..
What Bicyclists Are Supposed To Do
Bicyclists are supposed to abide by the same rights and responsibilities as car drivers.
We often times begin to think that bicyclists can just go where they may.
In California, it’s the law that bicyclists do these things:
- Obey all traffic signs
- Obey all traffic lights
- Ride in the same direction as traffic
- Signal when turning
- Signal when changing lanes
- Wear a helmet if under the age of 18
- Allow faster traffic to pass when safe
- Stay visible and not weave between parked cars
- Ride as near to the right curb as practical
- Do not ride on the sidewalk unless legal exceptions allowed
- Make left turns in the same way cars do
- Make right turns in the same way cars do
- At nighttime must have a front lamp
- Must have a rear red reflector or equivalent
- Reflectors on each pedal
Should we arrest every bike rider that does not obey the laws?
Imagine how many arrests you’d need to make. The jails would be filled with bike riders.
We can probably agree that we’re not going to be arresting all of these unlawful bike riders.
Unlawful Acts By Bike Riders
What kinds of unlawful acts am I referring to, you might ask, well consider these:
- Tend to ignore traffic signs and blow through stop signs
- Treat traffic lights as a game that regardless of light color try to get through unscathed
- Ride in the opposite direction of traffic (quite popular!)
- Never signal when turning
- Never signal when changing lanes
- Be nearly invisible and weave between parked cars
- Ride sometimes near the right curb but really wherever judgement suggests
- Ride on the sidewalk (often done to avoid wayward cars)
Car Driver Issues And Bike Riders
I’d like to next shift focus to the car drivers in the equation of bike riders on-the-road and mixing with cars.
There are some car drivers that outright hate bike riders.
These drivers believe that bicycles should be banned, or at least forced to only be used in say parks or at the beach, in places where no car traffic is allowed anyway.
There are many drivers that are happy to share the world’s roadways with bike riders.
Unfortunately, what often happens is a few bike riders cause a problem, and the car drivers take this out on all bike riders.
Sadly, there are also car drivers that seem to be living in their own bubble and rarely contemplate the plight of the bike rider.
For these blind-deaf-dumb car drivers, they don’t look for bike riders. They don’t anticipate what a bike rider might do.
AI Autonomous Cars And Bikes
What does this all have to do with AI self-driving driverless autonomous cars?
At the Cybernetic Self-Driving Car Institute, we are developing AI systems for self-driving cars and included is the development of specialized software related to bicyclists, which by some of the automakers and tech firms is considered an “edge” problem.
An edge problem is one that is outside the core of the overall problem being solved.
Getting a self-driving car to properly drive down a road, being able to stay within the lanes of traffic, make turns legally, and otherwise drive like a regular car is supposed to drive — that’s considered the core problem to be solved for AI self-driving cars. Having to deal with things like pedestrians, or things like bikes and bicyclists, well those are second fiddle and usually considered an edge problem. Definitely want to eventually solve an edge problem, but it’s not the highest priority.
We believe that solving the self-driving car aspects of detecting and avoiding hitting bicyclists is a crucial aspect of being on the public roadways.
A self-driving car that does not have provision for especially watching out for bike riders is about the same as the human driver that does not pay attention to bike riders. The head-in-the-sand approach will only last so long. Ultimately, inexorably, an AI self-driving car is going to hit a bike rider if there’s no particular capability in the AI to avoid doing so.
Some AI developers tell me that it’s easily solved.
If an object appears in front of the self-driving car, regardless of whether it is a bike rider or maybe a spaceship from Mars, all the AI has to do is detect the object and bring the car to a halt. It doesn’t matter that it’s a bike rider. The AI shouldn’t need to care. Any object, the rule is, don’t hit it.
Okay, I say, let’s follow that logic along. A child is riding their bike. It’s a school zone. The AI self-driving car is going the speed limit. We’ll say it’s going at 25 miles per hour (which is about 37 feet per second). The child, not paying attention to the car traffic, suddenly swerves in front of the self-driving car. The self-driving car needs to react. Can it come to a halt, having been going at 37 feet per second, in time to avoid the child that has nearly immediately appeared in front of the self-driving car? Answer, probably not.
Furthermore, maybe the self-driving car could have swerved to avoid hitting the bike.
Or, maybe the AI should have been anticipating that a child on a bike might make an erratic action, and so have gone slower, maybe decreased speed to 5 miles per hour, as a precaution.
Or changed lanes to give a wide berth for the bike rider.
Incorporating Bike Riding Elements Into The AI
A bike rider has certain characteristics that can be modeled and possibly predicted.
The bike and bike rider are not just any object.
They are not the same as a light pole or a fire hydrant.
They are usually a moving object, though can be at rest or stationary at times too.
They have a particular kind of profile.
We know that this moving object is intended to go in certain ways, and we also know that it can substantially decide to do something untoward.
I’ll walk you through the main elements of:
- Sensor Fusion
- Virtual World Model
- AI Action Plan
- Car Controls Command
The first aspect to consider is the sensor of the self-driving car.
The hope is to be able to detect the presence of the bike rider.
This can be potentially done via the visual sensors of the cameras. Imagine a picture of a street scene and you need to find the bike rider somewhere in the picture. This can be easy, if the bike rider is fully visible. This can be hard, if the bike rider is partially obscured by being behind another car or other objects. The visual aspects should be triangulated with the use of radar, sonar, and LIDAR (light and radar, if available on the self-driving car). Any of these sensors might catch a glimpse of a bike rider. The bike rider can appear and seemingly disappear, but hopefully at least one or more of the sensors is able to detect them.
Next is the sensor fusion.
This involves bringing together the sensory data and trying to reconcile it. The bike rider might be detected by the LIDAR, but the camera can’t spot him or her. Should the LIDAR be trusted or it is a false indication of a bike rider? The sensor fusion should be assessing which of the sensors is right or wrong, or at least potentially right or wrong. By combining together the bits and pieces from the multiple sensors, it possibly provides a strong indication of where the bike rider is.
During the virtual world model update, the AI should be tracking the bike rider.
Where did the bike rider initially get detected? How fast is the bike rider moving? Is the bike rider riding smoothly or erratically? Does the bike rider seem to be a child or an adult? Does the bike rider pose a threat to the self-driving car? Does the self-driving car pose a threat to the bike rider? What can be done to reduce the risks of colliding with the bike rider? And so on.
From the updates of the virtual world model, the AI action plan needs to get updated. Maybe the self-driving car should slow down, and so the AI will be instructing the car to do so. Or, maybe alert the bike rider that the car is nearby and a danger is ensuing, this could involve honking the horn or taking some other conspicuous action. Or, speed-up. Or change lanes. Etc.
Finally, the AI then needs to issue commands to the controls of the car.
This will then take time to be enacted. The AI will need to detect once the actual physical car has taken the action deemed needed, and then cycle back through each of these steps accordingly. In some cases, this will need to happen in split seconds and so the timing of detecting the bike rider, predicting their actions, updating the model, updating the AI action plan, and issuing the car control commands can be crucial to avoiding a collision.
Multiple Bikes At The Same Time
So far, the above highlights the acts of a solo bike rider.
In real life, the odds are that wherever there is one bike rider, there will likely be more.
It could be a school is nearby and a bunch of kids are riding their bikes to school. It could be a bike club and a gaggle of bike riders are out for their exercise. The point being that even though it seems like a hard problem to track and predict one bike rider, the odds are that this is a much more difficult problem because there are bound to be many bike riders all at once.
It becomes an interesting problem too to keep track of the various bike riders as though they are individuals.
Another factor to consider is daylight and nighttime.
Nighttime is going to be harder for the visual sensors of the self-driving car to detect a bike rider. Many bike riders do not have lights. This is a recipe for disaster. Inclement weather will also have an impact on the ability of the sensors to detect the bike rider. In short, the AI system cannot be programmed to simply assume that it will be nice and sunny, and that the profile of the bike rider will be one hundred percent noticeable.
There are also the use cases of a bike rider that is not actually riding their bike.
Perhaps the bike rider is walking their bike.
You might say this is then a pedestrian and no longer a bike rider.
I’d suggest that it is more of a grey area.
The walking person can suddenly hop onto the bike and start riding it.
The AI self-driving car should be anticipating this possibility.
The profile of a person riding a bike is also different looking than when riding a bike.
I mention the profile aspects because many of the AI self-driving cars use Machine Learning (ML) such as artificial neural networks for purposes of finding objects in visual images that are captured. The neural network is typically trained on thousands of pictures of people riding bikes. This then allows for the neural network to inspect a new image and try to gauge whether there is a bike rider in there. Suppose that the only pictures used to train the neural network consisted of riding bike riders. A walking bike rider then might not be detected as being a bike accompanied person.
For those of you further interested in this aspect of detecting a bike and someone walking the bike, you might want to read about my forensic analysis of the Uber self-driving car death in Arizona that involved a pedestrian walking a bike: https://aitrends.com/selfdrivingcars/initial-forensic-analysis/
Biketown versus bikelash.
Bicyclists, love them or hate them.
The AI self-driving car has to know about bicyclists since they exist and they are on the roadways. This edge problem is vital to becoming part of the capabilities of any proficient AI self-driving car. You could potentially have a Level 5 self-driving car that had no ability to detect and deal with bike riders (a Level 5 is considered the top of the scale and means that it is AI that can drive the car as a human can), but I would assert that such a lack in capability is not only a significant omission but I dare say not what we all would want a true self-driving car to be able to handle.
With a proficient AI self-driving car, there’s a fighting chance to reduce the 45,000 annual biker injuries and the 1,000 or so annual deaths.
Hold your breath for a moment when I say that if the AI isn’t good enough, we might actually end-up with more injured bike riders and more human bike rider deaths.
We cannot just assume that the AI self-driving car will magically eliminate those injuries and deaths.
Bikelash will become AI-lash, if AI self-driving cars start hitting bike riders.
Mark my words.
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://firstname.lastname@example.org
For Dr. Eliot’s books, see: https://www.amazon.com/author/lanceeliot
Copyright © 2019 Dr. Lance B. Eliot