AI Driven CArs
Introduction
Artificial Intelligence can be implemented within cars to make them fully operational without the use of a driver. This results in an enhanced passenger experience and greater safety for everyone.
Technology
Autonomous cars can operate without the use of a driver using Nvidia's gpus. AI in autonomous cars can navigate through obstacles using machine learning, perception technology, cameras, lidars, maps, ultrasonic sensors, and radars. These technologies have become increasingly advanced: lidars bounce lasers off the environment and use photons to identify speed; maps help to determine location accurate to the centimeter; perception technology can find obstacles 300 yards away. Together, these technologies can identify obstacles, signs, speed, and location. They use double processing because there are multiple of each technology and obstacles cannot be counted multiple times. They also have a hierarchy because some technologies can detect different objects at different times, so the hierarchy establishes when to believe certain combinations of technologies. Autonomous cars utilize multiple AIs and neural networks in tandem. AI is also implemented in non-autonomous cars: lane assist helps drivers stay in lanes or change lanes; ADAS (Advanced Driver Assistance System) utilizes AI for automatic braking, driver drowsiness detection, lane departure warning, parking and more; CarVi and Nauto's AI can evaluate drivers to help them get better.
Training
AI for autonomous cars goes through training and simulations with gpus, and using many gpus at the same time allows the AI to be trained in minutes. AI goes through thousands of simulations and trains with millions of labeled images. These images are of lanes, cars, pedestrians, traffic lights, bikes, and more. Finding commonalities in the labeled images gives the AI the ability to set probabilities to identify an object and create thresholds to determine an action. Sometimes, AI might not be able to identify an object, but it can be trained to go around unidentifiable objects. The specific training of the AI can determine the driving style of the car. For example, a car can be trained to safely and aggressively weave through traffic while using its horn and headlights.
Driving
In order to drive, the AI in the autonomous car goes through as 8-step process that happens 30-40 times in a single second. 1: Perception module - obtaining and processing the data from the sensors; 2: Multi-sensor - eliminating double counting from overlapping sensors; 3: Sensor fusion - going through the hierarchy of sensors if their data doesn't match; 4: 3D map - generating a 3D map of the environment; 5: using the 3D map to determine how the environment will change; 6: Localization - comparing the 3D map with an area map to help determine future actions; 7: Path planning direction planning - identifying the safest path to move on; 8: Actuation - controlling throttle, brakes, and steering (can vary depending on preference, as it can act as a sports car that stops and starts quickly or a luxury car that aims for a smooth ride).
Passenger Experience
AI in cars can interact with other cars and pedestrians to create a safer environment in order to prevent crashes and fatalities. Additionally, AI in cars can tailor the experience to its passengers by using cameras and other systems, such as Hyundai's READ. These systems allow AI to detect facial expression, heart rate, and electrodermal activity to understand emotion and expression in order to adjust to the passengers. AI can tailor the car to its passengers by changing the TV or radio channel and locking the doors when kids enter the car or slowing down when passengers are yelling. Without the equipment necessary for driving, the car's interior can also be customized. Inside the car, active noise cancellation and active suspension help to mitigate sounds and bumps. AI in active noise cancellation outputs noise to cancel out noise from the environment, which is identified using microphones. AI in active suspension surveys the road using sensors and adjusts the shocks and springs accordingly. Outside the car, AI can scan faces to ensure that the car is not stolen by an unknown person. Lastly, AI can be implemented in non-autonomous cars to entertain passengers and ensure that drivers stay awake.
