On one hand I understand that it is a monumental advance in technology.
On the other hand a computer can recognize a unique face from pretty far.
In: 8
Other people have given very good answers as to why this technological problem was so difficult to solve, but I’m surprised to see nobody pointing out that we have solved it. Self-driving cars are here, ready, and better than human drivers.
The reason they aren’t taking off is largely political– that doesn’t mean wrong or bad, just that they aren’t rooted in the technology anymore. As soon as we want to, we can.
First, You have to have the hardware to be able to see and interpret your environment. Basically, you need to be able to perceive where the road is and where a tree is. You need a way to tell if a car is coming at you and a way to recognize a pedestrian. You need to see road lines and you need to read signs to figure out speed or other info. Or have some other way to get that info in real time.
Then you need the software to take all those inputs from your sensors and actually figure out what they mean. Your software must be able to make the correct decisions day and night, in the sun, rain or snow. It needs to be able to adapt to both the rules of the road and to human drivers that don’t always follow those rules. It needs to be able figure out that a stop sign means stop… But a picture of a stop sign on a billboard next to the road is irrelevant. It needs to understand that a grocery bag blowing across the road cannot mean slam on the brakes, but an animal should. It needs to know that it must slow down for bicycles, but a bicycle on the back of a car is actually part of that car.
And a billion other wierd random things.
Self-driving cars are hard because of all the human drivers on the road. It’s very easy to program a car to follow all the rules. But humans don’t always follow every rule. People speed, change lanes without signaling, run through red lights and stop signs, etc.
If every car on the road was self-driven with no pedestrians, it would be a perfect system. But as soon as you introduce human drivers and pedestrians into the equation, things become exponentially more complicated.
Our current advancements in AI have been way overblown. Basically companies are hyping it up because they need investments to make it exist in the first place. They want us to think that we currently have the technology to start letting computers think through problems for us, but we’re actually still just feeding them tons of data and programming them to respond based on the data. This technology can do so many amazing things, but it’s not actually thinking or assessing its environment in any meaningful way.
Those self-driving cars you see videos of are actually loaded with tons of information added by the programmers about its surroundings. Basically it is told where the traffic signs, traffic lights, intersections, and etc are. So they are making some rudimentary decisions based on what it observes as it drives, but it’s helped Along by tons of manual work which wouldn’t be feasible beyond the small area of testing.
So to directly answer your question, self-driving cars are hard to make because we aren’t actually able to create computers that can analyze their environment and make decisions for themselves to that level. When you learn to drive you might feel like it’s kind of an art form and a weird dance between people and the rules of the road. It’s the kind of thing that humans are able to pick up, but it’s completely incalculable to robots at this time.
Because driving is extremely situational, so many things happen at the same time and the computer needs to be able to recognize each and every single one of them and act accordingly. **Our** automated response is quite good thanks for millions of years of training, that made us pretty good drivers over all, but computers don’t have such a thing.
The sheer amount of data and variables a car has to consider is astounding, some of which don’t even have a clear answer. Imagine you see a car coming directly towards you, counter traffic flow, you would swerve to the left, but the left’s counter traffic for **you**, but to the right is a car coming in traffic wise at 100 km/h, that’s not do much an if-then-else response, but rather a decision, each situation has nuance that you have to consider, good fast each factor is going, how much room and time you have to maneuver, what are the road markings and sections, how much of a priority is it to stick to them, a car has to ponder evasive maneuvers into counter traffic lanes, or passing maneuvers, against not entering counter traffic lands. How do you make a computer that literally **knows** which of those to do in each situation.
It’s a very complex situation and the few moments when breaking some rules is granted, you have to know which ones to break, and imparting that kind of nuance in a computer is amazingly complex
Latest Answers