Why I think an Autonomous Vehicle Won’t Be Able to Reliably Guess Where a Pedestrian Will Go

In 2016, Volvo ‘promised’ that we would have ‘Deathproof Cars’ by 2020. That’s next year! You may also recall that about this time last year, an Autonomous Uber Volvo crashed and killed a pedestrian. According to the NTSB the car spotted the pedestrian 6 seconds before it hit and killed her. Volvo then said that their system relied on the ‘driver’ who was watching a movie on the entertainment system to apply the brakes in an emergency situation. Is that reasonable when they are allowed to watch TV. I often don’t hear someone speaking to me when I am engrossed in something.

The insurance industry is very interested in risk with driverless cars, especially given the same brand, Volvo, said in 2015 that they would indemnify drivers of AV’s. That’s pretty cool because imagine if for the lifetime of the car you didn’t have to pay any insurance. That’s a pretty good saving on the premium you pay for these vehicles.

So back to pedestrians. While I’m waiting for approval for my back fusion surgery from ACC, I go for a daily walk. I have found a fairly flat walk to our local beach, which is part of my core fitness regime.

In New Zealand we drive on the left hand side of the road so you would think that pedestrians facing others would also make way for them by keeping left. Not so in my experience. Some think that way, but most just walk where they walk, it may be that they want to be farther away from traffic or they just don’t think about it at all.

In busy urban environments many solo walkers have developed a system where they can avoid bumping into other people. They often make eye contact, make an almost imperceptible move towards the left or right and monitor to see if the other person has likewise recognized this and moves the other way. Usually this works so well we are almost unaware of the communication.

It works a large percentage of the time, but if a hundred people are at a busy crossing, I bet that at least 2 people will do that St Vitus’ Dance where they both go to the left, then they both go to the right and then they stop in front of each other and apologise, like ants sharing information as they march to the food store.

On my walks I find people who are oblivious, deep in thought, or perhaps listening to something through their headphones or noise cancelling earbuds. They don’t move.

There is the Alpha Male (I have fun with them, maybe there is a little alpha in me too). They want to show their domination by deliberately not moving even after that eye connection. They want me to move, but I don’t. That confuses them because it works most of the time. I find that often if I look them firmly in the eye, they will grudgingly move at the last possible moment.

There are the directionally challenged people who can’t walk in a straight line.

There are people who suddenly change their mind. You see them all the time in supermarkets and shopping malls. They are walking in a certain direction as you walk behind them, then they suddenly turn around and walk smack into you, perhaps thinking that they are the only person in their little world.

In those malls you also have the person facing you who hasn’t moved for 5 seconds, pondering their next purchase. Suddenly they remember where they are and why, and march straight into your face.

The same happens with people crossing the road. I think I’ve seen everything. People step on the road and step off. They go half way across the road and change their minds. They go all the way over and change their minds. They stand and move as if they are going to step and don’t. They run across the road. They ‘silly walk’. They stop in the middle of a lane waiting for someone else.

I’m sure you get the point by now. I keep going back to Dan Ariely. People are predictably irrational. How do you train a car AI to understand how people will behave, when the people don’t understand themselves?

Marc Hoag of the Autonomous Cars podcast had an interesting thought about the insurance topic in Episode 83. There are more elements to the human problem. Humans programmed the computer algorithms. People installed the sensing equipment. Someone has to make sure that firmware upgrades are installed. I wonder if Volvo thought of those things when they made their ‘promise’?

So if humans will bump into each other more than 1% of the time and they create the ‘intelligence’ to stop cars bumping into each other; and in an autonomous vehicle crash, they still want to blame humans for the fatality (if I recall correctly they blamed both the pedestrian as a first reaction and then the hapless TV watching ‘driver’). Then there is a risk of human fault inherent in the system, and I haven’t even mentioned the people who built the computer or the LiDAR system as an OEM for the car manufacturer who installed it after having an argument with the boss, or the dense fog or ambient light that stopped it from sensing effectively.

Bottom line? There are many unresolved issues that need to be addressed before insurance companies are prepared to cover driverless cars. I wonder if they will consider them more risky than human driven cars for a time?

So humor me and try this exercise when you get the chance. When you see someone walking the other way on the pavement next. Walk on the same side of the pavement and don’t veer. See what happens. Of course don’t do it when they are supervising children and don’t in any way risk that anyone could hurt themselves. Most of the time we do have a pretty good human radar guidance system, but it takes two to tango and I bet you will do a dance with someone in the near future if you try this.

I welcome your comments.