If you think that which can drive itself is a part of a futuristic film, congratulations, you are on the wrong planet. Hardly can anyone be surprised to hear that cars which do need to be driven manually do exist. Numerous models have already been released and tested. The question still is what prevents them from reaching the market? The answer is simple: safety. There is still no guaranty that such an auto will not put your life and lives of other people in traffic on the edge of danger. While Tesla plans to partially implement such technologies in their future lineup and Google tests its new models of the type, we try to outline what self-driving cars need to improve.
Even though statistics says that driving U.S. is much safer than in most other regions, fatal crashes still occur every 3 million hours. It means that driverless autos need to show results much lower than this statistics which is quite difficult do. The culprit is software. Hardly can you remember a laptop, smartphone or any other device which can work for a long time without any bugs, crashes or freezes. Do you think such errors are acceptable in a car? They can be fatal. That is why current software needs significant improvements to be installed on an auto.
On Map Details
When self-driving models are tested, they operate with maps designed especially for a test drive. Every street and corner are already in the database. Thus, an auto knows where to go and where to take a turn. All it needs is to fill the map with obstacles. However, today’s GPS and other similar technologies may not run as smoothly in real time as pre-designed ones. So the car may not know where to ride to until the map is loaded. What is more, it may not sense pedestrians and other cars which can also end up in a tragedy.
Another point where such cars are at a loss is the use of sensors. They should know how to see the difference between dangerous and safe situations. Otherwise, it will hit the brakes all the time without a meaningful reason. For example, it is quite difficult to spot a nail on the way until you drive over it, while a paper bag rolling over a road can be a tricky object. These are the aspects which driverless cars still need to learn. They also need to understand how to estimate the time when a pedestrian is waiting to cross the road.
How can you teach a robot ethical aspect? There are situations when you need to decide which side to swerve. There may be a choice between killing a family of five people or killing one man on a motorcycle. Can you answer this question in a critical situation quickly? So what about a robot?