> "if it crashes no one dies"
That's a very optimistic assumption. A bike crashing into a pedestrian can definitely kill them if it knocks them over and they hit their head on something hard. Lots of elderly people even die of complications from broken hips. And a bike, even at low speed, can do a lot of damage to a small child.
A bike that swerves in front of a car can easily cause the car to lose control and crash into a person or another car, resulting in death or serious injury.
For these reasons, I'd expect that getting self-driving bikes to work safely (and to be allowed by regulators) would have most of the obstacles that self-driving cars face.
Note that in certain places (e.g., NYC), bicycles are subject to all the same traffic laws as cars. For example, they must stop at red lights, yield to pedestrians in crosswalks, respect one-way streets and not ride on sidewalks.[1] (It's not clear whether it would even be legal for an autonomous bike to drive in a bike lane, since it's certainly a motorized vehicle when driving autonomously.) Thus, an autonomous bicycle would have to have the navigation skills to be able to safely share streets with cars, trucks and buses in dense urban traffic. Can an autonomous bike carry around enough computing power and sensors to handle those kinds of situations?
[1] While most bicyclists in NYC ignore traffic laws, a self-driving bike that's not even capable of obeying such laws would certainly be frowned upon.
