Transportation

While there are several applications of AI/ML in transportation, we will discuss the most complex one which is Self-driving Vehicles. This is probably one of the most complex applications of AI/ML currenly being developed. Self-driving vehicles need AI/ML very close to human intelligence, learning and memory, since these are machines that can cause extensive injury to humans and damage to property if their algorithms fail. They use a combination of multiple algorithms and huge amounts of data both from static sources and from dynamic, real-time sources such as sensors.

A self-driving vehicle is a ground vehicle that is capable of sensing its environment and moving safely with little or no human input. There are two key components that a self-driving vehicle requires:

• A wide range of sensors to read data about their surroundings, such as cameras (to read images of signage and road markings), radar, lidar, or sonar (to detect obstacles), GPS (to determine the absolute location and get directions to the destination), odometry (to detect relative positioning) and inertial measurement units (including accelerometers and gyroscopes to detect the speed and orientation of the car).
• AI/ML systems take data from the sensors as input, process that data using the relevant algorithms, take decisions similar to what a human would, and send control signals to appropriate parts of the car to take action, similar to what a human driver would.

The AI/ML algorithms have to be extremely complex and mimic human behaviour as closely as possible including taking decisions based on critical thinking and reasoning.

• One of the most critical decisions the automated driving system has to take is when to stop to avoid a collision. It is not practical or safe for the car to be over-cautious and stop every time it believes there is an obstacle in its path. Depending on what the obstacle is, whether it is moving away from the impact itself and whether it's a dangerous situation such as a collision with another car or an object (in which case the car must stop immediately) or an obstacle such as a gate (in which car needs to slow down and roll to a stop rather than an immediate stop). All of this is more critical if self-driving cars are in the path of human-driven cars since humans apply rational thinking when taking a decision (every obstacle does not mean slamming on the brakes) and they expect cars in front and behind them to take similar rational decisions as well.
• The second critical decision is to follow road and traffic patterns and signs. The cars cannot change lanes at will, and if they see a stop sign or a red light, they must come to a gradual stop and start again at the appropriate times. Failure to stop could be dangerous and failure to start on time could cause inconvenience to drivers behind.
• Cars also have to follow speed limits which could either be pre-programmed into their systems with map data or they will have to read the speed limit signs just as humans do and adjust their speed accordingly.
• Cars must be able to navigate the roads and follow directions to get from the origin to the destination in an optimal manner.

This is not a comprehensive list of algorithms that are required, there are many more algorithms that are programmed into various components in a car to achieve the objective of a safe and practical self-driving car.

As you can imagine, machine learning and data both play a very important role in building efficient algorithms. The more the data the more patterns the programs can detect and the more accurate their decisions will be.