How self-driving
car technology works

A transportation revolution is happening now. Self-driving car technology, or Automated Drive Assist Systems (ADAS), in vehicles where a driver is present and can take over at any moment,  is and will continue to improve how we live, work and interact with each other. The benefits are enormous and manifold.

Though only about 36% percent of Americans drive each day during the pandemic, each motorist travels about 37 miles per day. Tens of thousands of Americans die each year in car crashes, and many of them are caused by human error. Self-driving cars are expected to greatly reduce those crashes. We could also realize increased fuel efficiency from autonomous vehicles, as well as the potential replacement of some drive and commute time with in-car leisure time.  

To answer the question as to how self-driving cars work, it's important to understand that the evolution toward fully self-driving cars is already well underway. The Society of Automotive Engineers (SAE International) categorizes vehicles into six levels of autonomy:

1. Warnings and Momentary Assistance - Audible and visible notifications are made to alert the driver regarding blindspots and lane departures, and automatic emergency braking.
2. Driver Assistance - Car assistance for isolated functions, like accelerating, braking and steering. Nearly all cars on the road fit this category.
3. Partial Automation - Assistance for two or more functions together, such as cruise control plus automatic braking. Luxury and technology-forward cars built within the last few years often match this classification.
4. Conditional Automation - Self-driving in specific situations but requires an attentive human driver at all times to take over when needed.
5. High automation - Self-driving car technology enables driving without a driver but only in some situations. Only experimental or limited-purpose cars match this classification at present.
6. Full automation - Fully autonomous in all conditions and scenarios. Class five vehicles do not need human controls, like steering wheels or brake pedals, and they rely entirely on self-driving car technology. This class of cars is still in the works and does not yet exist for passenger-bearing vehicles.

The fully autonomous cars of the future could be rolling supercomputers—artificially intelligent robots on wheels. Numerous sensors and onboard computers may be guided and assisted at all times by high-speed wireless networks and edge computing resources.

So, how do self-driving cars work? Here's a summary that begins to answer the question: Cameras, radar and lidar (light detection and ranging) capture a 3D, 360-degree picture of the surrounding environment. Computers isolate, interpret and identify objects, people and animals, tracking the motion of everything and making predictions about where anything that's moving will be from moment to moment.

For example, these cars would be able to identify a dog sitting by the side of the road, and they could also predict, calculate and factor in possible behavior. Self-driving cars may also be able to read bicyclist hand signals, road signs, street lights and anything painted on the road (such as road dividers, bike lanes, crosswalk markings and words, like "bus only"). This "knowledge" may be combined with other inputs, such as GPS location data, maps and traffic data, and it could result in a continuously updated "path plan" for where the car will drive, how fast it will go and when it will turn, stop and slow down.

Some information about a car's surroundings could come from external sources. For instance, cars encountering roadblocks, serious potholes, work zones, emergency situations and other types of disruptive events may communicate that information to other self-driving cars, emergency systems, smart city networks and roadside networks—this is also known as cellular vehicle-to-everything (C-V2X). Vehicle networks and other self-driving cars would receive that data and process it to affect vehicle decision-making.

Self-driving car technology might generate terabytes of complex data during each trip. Onboard computers would offload the processing of some unstructured data for analysis by special purpose machine-learning programs running in edge data centers. Passenger safety (and data security) would rely substantially on very low-latency 5G and mobile edge computing (MEC), enabling cars to make split-second decisions in near real time.

Smart cities may also broadcast their own data about traffic, detours, parking availability, emergency situations and other pertinent circumstances.