Here are the three challenges the company is currently facing with the cars, according to Business Insider:
Snow: The vehicles, like humans, largely rely on the use of visual sensors to navigate roads. But snow is reportedly problematic for Google’s vehicles because it can cover lane markers and other visual cues needed by the vehicle to continue driving safely.
New or Changing Roads: One of the advantages of Google’s automated cars is that they contain every road, highway and sidestreet documented in Google Maps. When a Google autonomous vehicle comes across a road not yet entered into its map system, however, the car can get lost. Someday, autonomous vehicles may be able to learn and improvise in such situations, but solving such a complex artificial intelligence problem may take some time.
Taking Human Direction: Google’s autonomous vehicles also have trouble navigating through areas where traffic is being directed by a human using hand signals, such as a construction zone or a crowded parking lot, according to Business Insider. Much of Google’s work in making its autonomous vehicles work well has been a matter of designing the car’s sensors to identifty signs and combine “knowledge” of maps and data with discrete sensory input. The problem of decoding human behavior and ambiguous hand gestures is a separate problem that engineers may need to overcome before Google’s cars are considered safe enough to enter mass distribution.
And according to an engineer Business Insider spoke with, none of these problems are insurmountable. The engineer's view, according to the article, is that self-driving technology will enter cars gradually and eventually progress toward the fully-automated electronic chauffeur.
