Tech Tuesday: How do self-driving cars work?
Roads filled with self-driving cars may stop existing solely in fantasies and become real, thanks to research innovations in the technology industry.
Advancements needed to produce self-driving cars have been moving quickly, and industry giants predict self-driving cars will soon be available to the public. Nissan, for example, announced it will produce self-driving cars by 2020, while Google is aiming to do so by 2018, according to robohub.org.
Designing a self-driving car is no small feat. The car has to be able to do everything a human driver does, such as navigating, determining its surroundings, predicting what will happen with the surroundings and what the car should do next, according to the website for Google’s Self-Driving Car Project.
To do this, the car uses a global positioning system (GPS), an inertial navigation system (INS) and a series of sensors, according to robohub.org. Information from the GPS and INS are used to position the vehicle while the sensors produce a three-dimensional image of the environment.
GPS works by having satellite signals sent to the vehicle to find the vehicle’s position and velocity. However, this signal can be jammed, the reading is not very precise and can give errors due to background noise and signal reflection, according to a paper submitted to the North Atlantic Treaty Organization.
INS uses accelerometers in the vehicle to determine its orientation while still capturing positional information. Using INS and GPS together not only provides redundancy but lets the vehicle continue to navigate when the GPS signal is jammed, decreases noise and gives more accurate information, according to the paper.
A control system is used to make navigation decisions based on the filtered information the vehicle receives from the sensors and positioning systems, according to the site.
Most control systems use designs that make decisions by producing and maintaining a map of their immediate world, and then using it to find the optimal route to the destination while avoiding obstacles such as pedestrians and construction, according to the site.
The path is then broken down into individual commands, controlling the actuators, which determines how the vehicle steers, accelerates and brakes. This whole process is repeated many times every second until the destination is reached, according to the site.
A map of the vehicle’s environment is produced by using sensors such as cameras and lasers. Laser Illuminating Detection and Ranging (LIDAR) shows the vehicle its surroundings by reflecting laser beams off of the objects around the vehicle to determine their distance and size, according to makeusof.com.
LIDAR is extremely accurate for mapping surroundings, but cannot give real time updates on the velocities of surrounding objects. For this reason, radar units are in the front and back of the vehicle, helping prevent the risk of accidents, according to the site.
Cameras are also present not only to provide redundancy, but give the vehicle a sweeping view of its surroundings. This helps provide information such as the dimensionality and depth of objects, according to the site.
The GPS and INS systems provide a broad view of the vehicle’s surroundings while these sensors provide a more in-depth perspective, according to the site. All of the information gathered by these systems is aggregated and used to produce a map.
Obstacles are categorized by the vehicle based on the obstacle’s characteristics, such as the number of wheels and velocity, which are then compared to a preexisting database of obstacles. Knowing the type of obstacle ahead helps determine how the vehicle should react to it, according to the site.
Knowing whether a motorcycle or pedestrian is entering the intersection is important to deciding how to react. The vehicle uses past, present and predicted future paths of all immediate obstacles to determine the best path to take, according to the site.
The vehicle plans its path by first determining a long-range path, such as a major highway to take. It then produces a series of shorter-range paths that can be taken, such as lane changes and making a turn, according to the site.
Safety is taken into consideration when making these decisions, ensuring that the vehicle is actually capable of completing its path given its speed and direction while also avoiding obstacles, according to the site.
The whole process of planning a path, removing unsafe paths and telling the actuators how to behave takes about 50 milliseconds, according to the site.
There are still advancements to be made, such as overcoming limitations in understanding road scenery, functioning in various weather conditions and driving through unstructured detours, such as accidents, according to the site.
Self-driving cars also require new laws to be put into place, as there are inconsistencies across the nation. Companies like Google are pushing for laws to allow self-driving cars to be legal, according to hg.org, a legal resource website.
While pushing for more accepting laws and making new advancements, Google is also proving the worth of self-driving cars, safely driving over 1.5 million miles, according to their website.
Harshel Patel is a School of Arts and Sciences sophomore majoring in molecular biology and biochemistry. He is the digital editor of The Daily Targum. He can be found on Twitter @harshel_p.