When it comes to autonomous vehicles, oftentimes you’ll hear about Mother Nature posing some of the biggest problems to the technology’s future success. But wobbly two-wheeled cyclists might be the most difficult task for robot cars to handle.
A story from IEEE lays out some of the struggles that come from trying to train an autonomous vehicle to detect a bicycle’s moves on the road. IEEE says, for instance, the Deep3DBox algorithm, a new 3D object detection method, identified 89 percent of cars during one benchmark test, but with bicycles, it only spotted 74 percent. The situation, however, is improving, slightly. From IEEE
[Deep3DBox contributor Jana] Košecká says commercial systems are delivering better results as developers gather massive proprietary datasets of road images with which to train their systems. And she says most demonstration vehicles augment their visual processing with laser-scanning (ie lidar) imagery and radar sensing, which help recognize bikes and their relative position even if they can’t help determine their orientation.
Further strides, meanwhile, are coming via high-definition maps such as Israel-based Mobileye’s Road Experience Management system. These maps offer computer vision algorithms a head start in identifying bikes, which stand out as anomalies from pre-recorded street views. Ford Motor says “highly detailed 3D maps” are at the core of the 70 self-driving test cars that it plans to have driving on roads this year.
Human beings are already struggling to grapple with the idea of automated vehicles carting them around, and that fear could be compounded, as AV technology currently struggles to deal with the quick-changing movements of a cyclist. Košecká told IEEE: “Bicycles are much less predictable than cars because it’s easier for them to make sudden turns or jump out of nowhere.”
And that brings into focus yet another challenge facing automakers who want to bring AVs to the market far sooner than what’s likely possible.