Motorcyclist Entanglement Avoidance for AI Self-Driving Cars


By Dr. Lance B. Eliot, the AI Trends Insider

Here’s some sobering statistics for you. In the United States, approximately 13% of all car related crash deaths are motorcyclists. That’s a sizable proportion.

Indeed, I would assert that all car drivers need to be aware of and anticipating motorcyclists.  Doing so is done to hopefully avoid getting entangled into a deadly embrace with a motorcycle, and spare too the lives of those riding on the motorcycle, along with avoiding correspondingly related car driver deaths and car occupant deaths.

On my own daily personal commute on the freeway, I see at least one downed motorcyclist on the average each day. That’s right, one per day. That’s nuts! Most of the time, the motorcyclist appears to be Okay and they are usually either laying on the freeway or getting up and trying to standup their motorcycle. Sometimes, sadly, I see the aftermath and there is a mangled motorcycle, there are ambulances, and a body draped with a tarp on the ground. Visually, this is all very striking and a constant reminder of the dangers of motorcycle riding.

For motorcyclists, the per miles traveled ratio for deaths is about 27 times that of cars. Simply stated, motorcycle drivers are taking big risks when they get onto the roads. They are riding on a vehicle that is considered less stable than a car. They are riding on a vehicle that is much less visible than a car. They are riding on a vehicle that lacks the same protections as an enclosed car.

By far, motorcyclists tend to be males. Of female motorcyclists that died in crashes, 61% were riding as a passenger on the motorcycle.  If you are interested in additional details about motorcycle accident statistics, take a look at the annual counts published by the National Highway Traffic Safety Administration (NHTSA).

What does all of this have to do with self-driving cars?

At the Cybernetic Self-Driving Car Institute, we are developing AI that helps self-driving cars to avoid getting entangled with motorcyclists.

You might already assume that a self-driving car should be watching out for motorcycles. Not so. The typical self-driving car right now assumes that a motorcycle will pretty much stay out of the way of the self-driving car. It is assumed that the self-preservation instincts of the motorcyclist would keep them from getting too close to a self-driving car.

This is the equivalent to a human driver that is unaware of motorcyclists around them while driving.

Regrettably, there are human drivers that are clueless when it comes to motorcycles and being on-guard about them. Those drivers have their head-in-the-sand and just assume that any motorcyclist stupid enough to get entangled with a car deserves what they get. Must be some kind of Darwinian process, these drivers assume. Focus on cars, rather than motorcycles, and besides how much damage can a motorcycle cause to their car. Just not worth the energy to think about motorcyclists. They are like fleas that happen to be around but no need to be defending against them.

What is perhaps even worse are the human drivers of cars that are out to get motorcyclists.

Let me explain that point. During my daily commute, I see motorcyclists that weave in and out of traffic, and manage to therefore avoid the doldrums of the inch-at-a-time bumper-to-bumper agony of us in cars. Believe it or not, I often see car drivers that appear to purposely try and place their car in the path of a motorcyclist. It is as though the car driver is angry about the motorcyclist being able to freely traverse the crowded roadway. These drivers seem to be saying to themselves, I’ll show that no good son-of-a-gun that they can’t just zip all around, I’ll put my several ton weighing car right in front of that motorcyclist and force them to slow down or altogether keep from getting in front of me. That’s the twisted thinking that seems to be happening with those envious car drivers.

For the moment, let’s shape a helpful framework about the mindset of the motorcyclists and the mindset of the car drivers.

There are car drivers that are normally observant and tend to be mildly wary of motorcyclists. These car drivers aren’t especially on the alert about watching for a motorcycle and only think about a motorcycle when one gets a bit close or otherwise draws attention. At least they aren’t ignoring motorcycles entirely.  We’ll refer to them as “normal” car drivers with respect to motorcycle awareness.

There are car drivers that are a bit crazy and try to get motorcyclists in trouble. We could almost include in this category the car drivers that are entirely unaware of motorcyclists, simply because those car drivers often make mistakes that appear to be an intentional act of trying to get a motorcyclist in trouble. Anyway, let’s just say that there are abnormal car drivers with respect to motorcycle awareness, and we’ll call them “crazy” for ease of reference.

We can somewhat do the same categorization for motorcyclists. There are some motorcyclists that are very careful about how they ride their motorcycle and are wary of car drivers, which we’ll call “normal” motorcyclists with respect to car awareness. But, there are also some motorcyclists that seem to want to taunt car drivers. Maybe these motorcyclists have gotten jaded because they’ve seen so many stupid car drivers, or maybe they’ve lumped over time the crazy car drivers into being all car drivers. Whatever way you see it, let’s just say these are abnormal motorcyclists that seem to almost have a death wish and make highly risk maneuvers in the midst of cars.  We’ll call them “crazy” motorcyclists for ease of reference.

Here’s what we have:

  •        D1 — Car drivers: Normal awareness about motorcyclists
  •        D2 — Car drivers: Crazy awareness about motorcyclists
  •        M1 — Motorcyclists: Normal awareness about car drivers
  •        M2 — Motorcyclists: Crazy awareness about car drivers

Next, imagine a 2-by-2 grid with these aspects. We’ll put the car driver categories on the vertical axis and put the motorcyclist categories on the horizontal axis.  This gives us four instances, consisting of D1:M1, D2:M1, D1:M2, D2:M2.

Here’s my take on these intersecting categories.

D1:M1 (Car driver normal awareness, Motorcyclist normal awareness), which I bet is likely the lowest of the deaths rates since this has car drivers watching out for motorcyclists, and motorcyclists watching out for car drivers.

D2: M1 (Car driver crazy, Motorcyclist normal awareness), which I am betting is one of the highest death rates since this consists of car drivers that are purposely trying to harm motorcyclists.

D1:M2 (Car driver normal awareness, Motorcyclist crazy), which I am betting is another high death rate since the motorcyclist is purposely putting themselves into harm’s way.

D2: M2 (Car driver crazy, Motorcyclist crazy), which has got to be the pinnacle of motorcycle deaths since you have the craziness of a car driver that wants to get motorcyclists that is then sparked by the added craziness of a motorcyclist with a death wish.

I would hope that self-driving car makers get past being a D1 mentality when it comes to the AI of the self-driving car. We need self-driving cars to take that extra step and be especially mindful of motorcyclists. No head-in-the-sand self-respecting self-driving car should be allowed on the roadways.

At least we ought to ensure that no self-driving car becomes a D2.

Why would a self-driving car be acting like a D2? It could happen by the oddball aspect that if a self-driving car is not equipped to detect and avoid motorcyclists it will potentially allow some other rules of the AI to cause it to misstep into the way of a motorcyclist. Keep in mind that the self-driving car is reacting to other cars. If other cars are seemingly taking evasive actions due to a motorcyclist, but if the self-driving car is clueless about why those cars are taking those evasive maneuvers, the AI of the self-driving car could inadvertently react to those other cars in a manner that it then becomes essentially a D2 that directly steps into the path of a wayward motorcyclist.

A self-driving car should be using its sensors to detect motorcycles. The AI then needs to anticipate what the motorcyclist will do. This anticipation then needs to be turned into action. Our AI component for motorcycle entanglement avoidance has embodied within it the various approaches that motorcyclists often take.

Here’s some examples.

Emergency Lanes. Motorcyclists will sometimes go into an emergency lane to quickly skirt around car traffic. Whether this is legal or not is a moot point. It happens. The self-driving car should be watching for this behavior. By doing so, the self-driving car will be aware that for example if it needs to use the emergency lane for a true reason, there might be a motorcycle that will block it. Or, if the emergency lane already has something in it, like a stalled car, the motorcyclist that is getting into the emergency lane might not be aware of this upcoming issue and so at the last minute try to swerve back into the way of the self-driving car.

HOV Lanes. Motorcyclists will often be in an HOV lane. This is usually legal and indeed by design a legal aspect since the viewpoint of society is that the motorcycle is a less polluting vehicle. But, where things go amiss is when the motorcyclists decide to dart into and out of the HOV. Whether this is legal or not is a moot point. It happens. A self-driving car needs to be aware that a motorcycle in a HOV lane might at any moment wander out of it. Or, a motorcycle might decide to dart across lanes of traffic to get into an HOV, even if not at the appropriate place to do so.

Lane Splitting. Motorcyclists in many states are allowed to split lanes. This means that they can go between cars and make their way without being quite so obvious. Self-driving cars need to anticipate that a motorcyclist is not bound by the normal conventions of a car. A motorcycle can at any moment zoom right along and narrowly skirt past the self-driving car.

Front of Intersections. Motorcyclists will often opt to meander through stopped car traffic, such as at a red light, and make their way to the front of the pack. A self-driving car that was at the front of the line, and awaiting a green light in order to proceed, now might suddenly have a motorcyclist directly in front of the car. This can happen with sufficient time to detect and thus prevent the self-driving car from rocketing forward once the light goes green (and avoid hitting the motorcyclist), or it can sometimes be a razor thin time margin if the motorcyclist zips up to the front just as the light is about to go green.

Rounding of Right Turns. Motorcyclists will sometimes try to go around a car that is stopped and waiting to make a right turn. The car might be waiting for pedestrians to finish crossing the crosswalk, or maybe waiting for car traffic to subside and then make the right turn. There are motorcyclists that come upon this situation and figure they will swing wide around the car and make the right turn. This is highly dangerous, and normally illegal, but anyway it does happen and enough that a self-driving car needs to be watching for it.

Stop Sign Rolling Thru. Motorcyclists will sometimes do a rolling stop at a stop sign. Rather than coming to a complete halt, the motorcyclist figures they’ll kind of cruise through the stop. A self-driving car that is expecting any vehicle to come to a full stop will often get caught off-guard and thus the chances of an entanglement with the motorcycle is heightened.

When designing this specialized AI component, the approach consists of anticipating the “crazy” maneuvers that some motorcyclists make. I realize this might seem like I am suggesting that motorcyclists drive illegally or otherwise are a danger to the roads. I am not saying this. I am merely pointing out that a self-driving car should be ready for all contingencies, including motorcycles that drive normally and ones that don’t.

We have devised two major approaches to the AI doing something about these anticipatory practices. There are actions by the self-driving car that we consider to be passive avoidance, while other more overt actions are known as aggressive avoidance. Let’s explore these two approaches.

Suppose the self-driving car is driving on the freeway during rush hour. Cars are moving slowly. A motorcycle is coming from behind of the self-driving car. As it does so, it is lane splitting, and on the left side of the self-driving car. Human drivers, when detecting this, will often move over in their lane toward the right edge of the lane. This then provides a greater space for the motorcyclist to pass along the left of the car. In fact, some motorcyclists look for this behavior and even at times give a thumbs-up to those thoughtful car drivers that provide that added space.

The self-driving car is not especially seeking to get a thumbs-up per se, but instead trying to minimize the odds of an entanglement with the motorcycle. Thus, if safe to do so, the act of moving over to the right edge of the lane would increase space for the motorcycle as it passes along, and reduce the risk of the motorcyclist hitting the self-driving car or otherwise encountering some other malady. Notice that I mentioned that it has to be safe for the self-driving car to do this. If the self-driving car gets overly close to the right edge of its lane, and if other cars in the next lane aren’t paying attention, it can cause one of those cars to adversely react, and the next thing you know there is a car accident in a domino kind of reaction.

An aggressive action to the above-mentioned aspect of avoiding the lane splitting motorcyclist might consist of having the AI of the self-driving car opt to change lanes entirely. In other words, rather than simply nudging out of the way, the AI might decide that the motorcyclist seems bent on danger and so want to move further away from the motorcyclist. This might necessitate making a lane change, even though there might not have been any other reason to make that lane change. The aggressive avoidance move makes bolder steps to avoid an entanglement.

Our self-driving car AI component scans for motorcyclists on a nearby basis and a faraway basis.  The faraway basis keeps track of motorcycles that are behind the self-driving car at a distance, and that might eventually make their way up to the self-driving car. Likewise, the self-driving car is tracking motorcycles that are a distance up ahead of the self-driving car. The self-driving car might ultimately catch-up with those motorcyclists and so needs to be anticipating the actions of those motorcycles.

The nearby basis detection deals with motorcycles once they get within a few cars lengths of the self-driving car. This is when the odds of an entanglement are at their greatest. If the self-driving car has been able to detect the motorcycle when it was at a faraway position, the self-driving car is more likely to be ready for a nearby action by the motorcyclist that could be troublesome.

Detecting motorcycles is not easy. Unlike cars, a motorcycle has a much slimmer profile. It can be readily hidden from view. Once it is spotted, it can then seemingly disappear from view. All of the sensors need to be used, including sensors such as the camera, the LIDAR, and radar, the sonar, etc.

Detection and tracking is actually a hard problem. The detection and tracking can be enhanced by using the anticipatory AI elements. If the AI anticipates that a motorcycle, once spotted, but now let’s say it is on the other side of a large truck in the lane over, will likely reappear to the right of the truck. Thus, the sensor data can make more sense even when only small clues exist as to whether a motorcycle is present or not.

So far, I’ve been discussing motorcycles as though they only appear one at a time. Of course, in any driving situation, there are bound to be multiple motorcycles involved. Often, motorcyclists like to travel in packs, under the belief that it is perhaps safer and they are more readily seen by car drivers. Plus, motorcyclist often have an alliance among other motorcyclists, and will at times simply “join up” when they spot each other in a given driving situation. Therefore, the self-driving car AI needs to be tracking multiple motorcycles at once. And, there are at times group behaviors of motorcyclists that are important to be aware of.

Based on movies and TV, we have a societal image of the motorcyclist as being a maverick. They travel the open roads. They take no gruff from anyone. They are free and easy riders. Maybe, but from the perspective of a self-driving car, they are an object on the roadway that is in motion and can be a danger to the self-driving car. The self-driving car needs to protect itself, and also help the motorcyclist by doing whatever the AI can do to keep the motorcycle from getting into a dangerous condition. It is a mutually beneficial relationship in that sense. Avoiding entanglements cannot be left to chance. The AI for a self-driving car needs to include a distinct ability to cope with motorcycles. That’s a must for true self-driving cars.

This content is originally posted on AI Trends.