Skip to main content
  1. Reviews/


·5 mins

Driverless cars are autonomous robots which drive cars with human-like precision, call them robotic chauffeurs if you will. Imagine a world where cars drive themselves, pick you up from any place, and then leave you to your destination; and then imagine a fleet of cars coordinating among themselves transporting materials and people without any human intervention. These cars take the shortest path considering road conditions, traffic situation and any other disruptions by being preemptively notified. Driverless cars or Self Driving cars which caught our fantasy few years ago are a reality now. Google was one of the first companies to build a prototype and iterate over the prototype for years. Hod Lipson and Melba Kurman’s “Driverless” is an ethnographical and anthropological study which traces the evolution of Self driving cars from being a public fantasy to a fully realized industrial grade prototypes and models.

The idea of self driving cars or at least autonomous transportation has caught public imagination for more than four decades. In the early part of 20th century a small scale version of autonomous transportation was on display which caught everyone’s fantasy in the day. One of the biggest missteps towards autonomous transportation was the work towards intelligent highways which regulate and automate the transportation of vehicles in its lanes. According to the authors, the idea was doomed from the start because it did not leverage existing infrastructure and needed special vehicles to run on the highways; which was, let’s face it, not going to happen.

Brian Arthur in his Nature of Technology illustrates how each technology is a recursive application of other technology. Similarly, the technologies which enable self driving car came to fruition with advances in Robotics, Mechanical Engineering, Hardware and Artificial Intelligence. Let’s take apart the self driving cars and see how it works. Self driving car is a robot, and like other robots it has sensors for instrumenting its environment and actuators, its way to react with the environment. The actuators are, of course, the constituents of car like brakes, accelerators et al. A self driving cars uses a plethora of sensors to make sense of its environment. LIDARs, RADARs and SONARs are used to estimate distances between the car and other hurdles. GPS receiver is used for determining current position of the car and its distance from the destination. Inertial Measurement Unit or IMU for short is used for measuring the vehicle’s inertia and direction. The single most important breakthrough which rapidly evolved the self driving car prototype to workable implementation was - Deep Learning. Deep Learning is an up and coming field in Machine Learning Research which uses Neural Networks to extract patterns from a data set to automate any kind of decision making. Hard coding rules in the software which drives the self driving car is not a scalable solution. The deluge of data from the Internet has enabled companies to develop Deep Learning software which identifies any hurdle or roadblock with a great degree of accuracy. With Reinforcement Learning in the picture as well, we can have a software program which learns to drive by learning and making mistakes. As a matter of fact, hobbyists can and are making use of these cheap parts and modifying their cars so that they drive themselves or increase automation. As of this writing, an open source repository, openpilot, claims that it is at par with Tesla’s autopilot software and can be installed on some models of cars from Honda. The New York Times carried out an article covering how the self driving cars work.

Trolley problem is the ethical framework that the authors used to analyze ethical problems faced when we undertake widespread adoption of self driving cars. Imagine you are driving a vehicle and you come across a fork in the road or the tracks. In one the tracks, five people are tied to the tracks and it is invetible that these people will die when you take this track. In the other track, there’s only one person tied to the track. You have to choose one of these options, what would your choice be? This is the simplest form of the trolley problem. There are other variations as well. In a variation of the trolley problem, the one person tied to the track has greater intrinsic social value, say the person is a doctor, what would the choice be then? In another variation the person tied to the track is the driver’s son, someone who has more value to the person in control, what should the choice be then? There’s an online platform called the Moral Machine which is using crowdsourcing to understand ethical decision making in the large using the trolley problem. Autonomous vehicles, and by that I mean the people in charge of making them as well as the Government machinery which allows their usage, are continuously studying the challenges and ethical repercussions of the decisions. Whether these decisions will be outsourced at run time to a Moral Machine, only time will tell. With increased profiling of the people using the Social Machine, I will not be surprised if such a machine or algorithm already exists.

The sociological ramifications of a self driving car culture are also interesting. Owning a car, at least in America, is a rite of passage. The car manufactures have harvested people’s aspirations and designed cars for people which have varied tastes. How is self driving cars going to change that? The ride-sharing services like Uber have already made much of the driving redundant. Self driving cars operating in an Uber-like fashion might become the transportation du jour of the future. There’s a long laundry list of cui bono candidates using self driving cars - advertising companies, software sellers, hardware sellers etc. On the other hand, there are clear losers or at least it appears like that - Truck drivers, existing car companies. Self driving cars are boon for urban transportation and fleet management, but any technology that succeeds in the long run should be Pareto efficient or at least zero sum.

All in all, I found the book to be fascinating and insightful. Two thumbs up.