A World Without Stoplights?
Applications of AI and Data Science
Choreographed Intersections Utilizing Autonomous Vehicles and Pedestrians
The goals of this smart system is to coordinate the vehicles of a city at intersections to move without collisions, resulting a choreographed “dance” of passing vehicles that do not stop, but merely pass without accidents. This performance is measured by the expediency of vehicles to their destination, accuracy of crossing, and overall efficiency of fuel economy on all vehicles.
To improve performance of this smart system over time, knowledge collected should involve collision probabilities of vehicles crossings as well as braking or changes in speed of vehicles. This knowledge can be learned through a variety of ways: through experimentation and testing through simulations that the AI runs through (it is best to do testing through simulations as testing actual vehicles and human lives is not ethical), and data from existing traffic apps such as Google Maps and Waze.
In order to perform this task, the traffic AI must analyze vectors from other vehicles that are probable to cross the next intersection. When these vehicles are near intersection, the traffic controller AI takes over the vehicle from the vehicle’s onboard AI, to control speed and steering. The vehicles the either speeds up or slows down so that with all other approaching vehicles, all may pass without hitting one another. After successful crossing, the traffic controller AI then returns control of the vehicle to onboard AI or human operator. The human operator is notified of approaching intersection. The traffic AI and vehicle onboard AI does the rest; however, this switching of tasks between the two machines are seamless.
Potential data sources that should be tapped for this may come from simulations and tests made during the testing stage, satellite imagery data, and vehicle GPS. It may also come from cell phone location data of nearby pedestrians, in order to keep their safety. In addition, data can come from car make and model, and furthermore, the benchmarks associated with car models. Despite the complexity of this type of problem, the useful information that comes from these data sources are presented in a way that is quite easy to interpret. Data from simulations and initial tests may show as organized data since they come from a controlled source. Sources such as longitude/latitude, location data, velocity, distance from intersection, etc, are all quantitative data that can be easily processed by a machine. Similarly, information about each vehicle such as make/model and year are readily available as are benchmarks for these vehicles such as fuel economy. However, data such as the amount of wear on the vehicle breaks may be more difficult to interpret.
Division of Labor
These tasks can be decomposed into subtasks that are to monitor intersections and plot vectors of moving vehicles and pedestrians, to decide which cars to speed up and which to slow down in order that they can pass without collision, and to find the optimal time of takeover from traffic AI to onboard AI. Within these processes, computers should be tasked with deciding speed and plotting collision avoidance. Humans should plot destination, and tell computer car behavior in intersection — are we going straight or turning?