Most frequently people speak about Tesla Autopilot from driver’s comfort perspective since it allows drivers to be less tired during long commutes, even though it still requires human supervision while driving. But probably the most important aspect of the Autopilot is actually being the least mentioned - it’s safety.
AI doesn’t make human-like mistakes. It’s the reason why we are not amazed with the fact that Autopilot gets into accidents not so often as humans do. Though the pedestrians’ safety is yet another Autopilot’s advantage being unreachable for human drivers — its cameras and sensors see people crossing the road in such conditions when human eye can’t do just anything.
Andrey Karpathy, Tesla’s head of AI and computer vision, released the following video gallery at the Scaled Machine Learning Conference in February. It shows a few cases when the AI’s reaction prevents tragedies:
Karpathy mentioned that the Autopilot may not been activated in all of those cases. Though its sensors permanently monitor the environment around the car and may activate the emergency braking if necessary. Andrey added that Tesla AI development team see tens to hundreds of such videos every day and a good fraction of those accidents ends posivitely.
