Friday, 24 January 2020

Self Driving Cars


Imagine a world where you wake up, grab your cup of coffee, and hop in your car to drive to work, except you are not doing the driving. You have more time to sleep, read a book, or even get a physical. This is a world we all want to live in. And although we are not quite there yet, people all over the world are working on developing, testing, and planning for a future with autonomous vehicles. Because who doesn’t want that extra hour of sleep. So, Today in this tech blog we will find out the three most important challenges in building a Self Driving Car.

First is building a safe system without human intervention. But we can’t guarantee that accidents won’t occur as Murphy’s Law is always in play. Murphy’s law is an adage (important truth derived from experience) that states that “Anything that can go wrong will go wrong.” In order to replace human drivers, the self driving cars needs to be safer than a human driver. That includes human being stupid and crashing while drunk or looking at their phone. The Global safety report on road safety stated that in 2013 more than 231,000 people were killed in road accidents in India every year.

We can probably hold our vehicle to a higher standard. But that can be our benchmark for now. It needs to fail less than once every 231,000 times. Most of the vehicles nowadays are using lidar sensor which do not need light to see. The disadvantage of using lidar pulses is that it can be effected by heavy rains or low hanging clouds because of the effects of refraction.

Tesla unveiled its new purpose built computer, a chip specially designed for running a neural network, which Elon stated was one of its kind. It looks something like this

The fusion of sensors in Tesla’s Model 3 looks something like this

It doesn’t rely only upon visual camera’s but also makes use of 12 ultrasonic sensors which provide a 360 degree picture of the immediate area around the vehicle, and 1 forward facing radar.

Other manufactures have millions of kms driven to gain data but Tesla has over a billion. For training a neural network data is key. I will not go into the details of the neural network but the key take away for you is that the more data you have to train a neural network, the better it’s going to be.

Second challenge is building an affordable system that the average person will be willing to buy. The biggest disadvantage of the lidar sensors is that they are bulky and expensive. Tesla is focused on building not just a cost effective vehicle but a good looking vehicle.

Tesla’s mission to accelerate the world’s transition to sustainable transport has pushed tesla towards a cheaper sensor fusion setup. That is a cost effective solution and according to Tesla their hardware is already capable of allowing the vehicle to self drive. Now they just need to continue improving on the software and Tesla is in a fantastic position to make it work.

The third and the most important challenge is programming the vehicle on how to handle every scenario. This is a vital part of building not only a self vehicle, but a practical self driving vehicle. Programming for safety and practicality often conflict with each other.

If we program a vehicle for only safety, it’s safest option is not to drive. Driving is an inherently dangerous operation, and programming for the multiple scenario that can arise while driving is an insanely difficult task. Sometimes the computer will need to make difficult decisions, and may at times need to make a decision that endangers the life of it’s occupants or people outside of the vehicle.

It’s easy to say “follow the rules of the road and you will do fine” but the problem is if everybody obeys it why we need to have traffic police at each intersection. But if we continue to improve technology we could start to see road death plumment, while making taxi services drastically cheaper and freeing many people from the financial burden of purchasing a vehicle.         
  

No comments:

Post a Comment