Self-Driving Cars and Kangaroos and Localization
David Silver
413

Human drivers have to learn the best response for each animal type. But of course many human drivers don’t know this or make mistakes. We too have to learn the best tactic for the animals in each area. So it is not suprizing that self driving cars have to learn this too. There is no all purpose best solution.

An examle are deer on a mountain road. A common mistake is to steer the car to avoid the deer. but deer have an unpredictable behaviour. They will at first “freeze” and not move from the traffic lane and then as you approach closer will bolt to either direction unpredictably. This behavior saves them from predators but not from cars. So if the driver changes lanes to avoid the deer, the deer is just as likely to change into the same lane and be struck by the car. On a mountain road the best thing a driver can do is stay in the lane and apply the brakes for maximum effect and hope the deer bolts in one direction or the other. Changing lanes is MUCH more dangerous. But for other large animals such as a bear, the bear will quickly take the best/shortest path to avoid the car. Cats have different behaviour.

So unfortunatly for the designers of self driving cars, I’m afraid chosing the BEST maneuver might involve identifing the species of animal. Yes there is likely a default maneuver that is good enough.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.