It happened so fast, I had no time to react. I was headed home on my bike when a pickup truck came roaring past on the left, avoiding me by mere millimeters. It wasn鈥檛 any skill on my part that prevented a painful accident. It was pure luck.
If Google has its way, luck won鈥檛 be a factor in these all too common car vs. bike showdowns in the near future. Much has been written about , which replace drivers with sensors, servos, and many lines of computer code鈥攕ome of which was designed specifically with roadies in mind.
鈥淭he way we approached the problem is, 鈥榃hat鈥檚 the right, safe thing to do in each one of the cases?鈥欌 says Nathaniel Fairfield, a principal engineer at Google X. 鈥淲e look at a scenario or a class of scenarios, and we get a lot of data and a lot of experience. We look at how people have been interacting. What鈥檚 your instinct? What would I do in this case? And we also look at how different behaviors and approaches can work out.鈥
The main goal () is to avoid collisions with cars, bikes, and . 鈥淚n most cases, you really can get ahead of the problem. Instead of getting into a situation where you suddenly have to make a call about which way to go, you can anticipate the situation.鈥 That means the autonomous car needs to sense oncoming cyclists well ahead of time, and then determine whether it can clear the rider long before a possible collision.
To do this, the car has cameras, LIDAR, and radar that give it a 360-degree view of its surroundings. The cameras distinguish between things like stoplights and road signs; the lasers and radar determine the speed and direction of moving objects.
While the majority of the programming is designed to track any moving object鈥攂e it a car, bicyclist, or pedestrian鈥攖here is also some bike-specific code on board. 鈥淪omething we have found to be important are the hand signals,鈥 says Fairfield. 鈥淲hen a cyclist sticks out their arm, they are not just doing it for fun or accidentally. It is a very clear, very strong signal, so we have taught the car to recognize cyclists鈥櫬爃and signals.鈥
The car also knows what it means when a rider claims the whole lane. 鈥淸The cyclist is signaling that he] doesn鈥檛 want to be passed right now,鈥 says Fairfield. 鈥淔or a car driver, that鈥檚 sort of a judgment call: 鈥業s the cyclist trying to claim the lane, or should I blast past?鈥 That鈥檚 something that we are very responsive to. We really pay attention to when other road users are acting in a way that is sending us a message about what their intentions are.鈥
The Google engineers have also given cyclists the right of way at stoplights. 鈥淲hen we are stopping at a stoplight and a cyclist pulls up right next to us, we try to figure out what we should do. Should we go first, or should [the cyclist] go first? We can afford to wait and be conservative. When the light turns green, let [the cyclist] make the first move,鈥 says Fairfield.
Of course, the Google self-driving car isn鈥檛 subject to road rage, either. 鈥淲e are not sitting in that car feeling that adrenaline, getting irritated or angry. We are sitting back here, days later, looking at all this data and trying to understand what the appropriate response is and making a cool, collected data-driven judgment call,鈥 says Fairfield.
In the end, it boils down to one thing cyclists have been waiting to hear from car drivers for a long, long time: cyclists, like pedestrians, are particularly vulnerable road users. Google wants to be extra careful around them. And that could make all the difference.