We’ve been hearing about Google’s self-driving cars getting into fender-benders before, mostly at the fault of other vehicles with a human behind the wheel, but recently one of the autonomous vehicles got a bit confused by a cyclist at an intersection. Did the car not recognize the cyclist and almost collide with him? Nope. Turns out the rider was simply doing a track stand — where they keep the bike upright at a stop without taking their feel off the pedals — and the car couldn’t tell if they were moving or not.
Fortunately no one was hurt — there wasn’t even a collision — and the whole thing turned out to be pretty funny to everyone involved. What happened was that the car and cyclist were both stopped at an intersection, with Google’s ride having the right of way. The cyclist performed a track stand while waiting for the car to go, but his small movements to keep balance kept being detected by the car, causing it to start to move forward and suddenly brake over and over again.
During a track stand, as seen above, a cyclist will often shift forward and back in small movements to balance without needing to put their feet on the ground. In addition to these small movements, what likely contributed to the car’s confusion is the fact that during a track stand, the cyclist’s body position is basically the same as when they’re in motion.
In detailing the experience on a bike forum, the cyclist mentioned how the car’s pattern of starting to move forward, abruptly stopping, followed by a pause for a few seconds before trying again played out for about two minutes, with the vehicle never really making it past the middle of the intersection. Interestingly enough, the rider said he also felt safer with the self-driving car than with a human-operated one.
The two Google employees who were in the car also seemed to have found the situation humorous, as they were described as laughing during the situation and entering the data into laptops. The whole thing is actually a positive experience for Google’s autonomous car research, as the company stated how this kind of real-world feedback is needed to improve the software and algorithms that help the car understand its surrounding environment.