What is the technology behind autonomous vehicles?
Waymo’s fully self-driving vehicles’ latest hardware suite features vision, Light Detection and Ranging (LiDAR) and radar systems as well as supplemental sensors that, combined, give the vehicles a 360-degree view of the surrounding area and are designed to respond to objects up to 500 meters away. The hardware suite’s systems and sensors are instrumental in assessing and navigating roadways. The sensor suite allows the cars to create a detailed 3D map of the world in which they travel. Vehicles then use the same sensors to localize objects they detect while traveling. The LiDAR system, for example, beams out millions of laser pulses per second 360 degrees around the vehicle, measuring how long it takes for the pulses to reflect off a surface and return to the vehicle provides additional information about the car’s surroundings. The car then cross-checks this and other real-time information with frequently updated maps providing knowledge about the world to help the self-driving vehicle make smart driving decisions.
Cameras and LiDAR are used closely together to help the vehicle understand its surroundings. Cameras provide uniquely complementary information for objects, such as to detect colors, and help spot traffic lights, school buses, emergency vehicles, construction zones, and more. Radar complements LiDAR and cameras with its unique capabilities in weather conditions such as rain, fog, and snow. Supplemental sensors, including audio detection systems and a positioning system, are used to detect police and emergency vehicle sirens up to hundreds of feet away and provide information about the location of self-driving cars, respectively. Learn more about how self-driving cars work.
How do self-driving cars use data?
Self-driving vehicles, like Waymo’s, use data from previous real-world experiences, structured testing, and simulation, combined with real-time data from the cars’ suite of sensors, to help make informed driving decisions in the moment. Data continually feeds back into improving self-driving technology. For example, using self-driving data, engineers can identify challenging situations self-driving cars encounter and then decide to turn those situations into practice scenarios in simulation and on closed test tracks.
Waymo’s simulations are built using virtual replicas of roadways, including dimensions, lanes, curbs, and traffic lights. Engineers focus on replicating especially challenging interactions—for example, a hard to navigate intersection. Other variables are then added to the simulation, such as vehicles driving at different speeds, pedestrians, the timing of traffic lights, and more. The software then practices navigating these scenarios thousands of times and in different driving conditions.
How do self-driving cars see traffic lights?
Self-driving cars use a host of sensors that are continuously reading traffic patterns and scanning for moving and still objects, including traffic lights. High-resolution cameras can detect colors, allowing vehicles to understand if a light is red, yellow, or green. The cars know what these colors mean, which helps them understand the situation and proceed appropriately.
In addition to identifying and understanding traffic lights, Waymo’s self-driving cars are designed to identify and differentiate other road users, including cars, motorcycles, pedestrians, cyclists, construction work and other roadway activity. Fully self-driving cars can use the information they see to anticipate what other road users might do next and make the most-appropriate driving decision.
How do self-driving cars make decisions?
Predicting the behaviors of drivers, pedestrians, cyclists, and others on the road can be difficult. Machine learning models can help self-driving cars learn new and different types of behavior so they can make smart driving decisions. The sensor data and maps that self-driving cars employ are instrumental for navigation, but driving patterns vary across locations and road users may disobey traffic laws. This variety of data is what allows self-driving vehicles to more accurately make predictions that impact real-time decision making.
Waymo researchers have developed a new model called VectorNet to provide more accurate behavior predictions. Sensor data is simplified to points, polygons, and curves, which is then processed by VectorNet to capture the relationships between various vectors—for example, a car entering an intersection or a pedestrian using a crosswalk. Continually learning about these interactions allows the Waymo Driver to better predict other road users’ behaviors in a real-world environment and make better driving decisions.
Learn more about how VectorNet enables the Waymo Driver.
Can self-driving cars drive in different kinds of weather, such as snow?
Waymo’s self-driving vehicles undergo testing in extreme weather conditions, including snow and cold temperatures.
Waymo engineers also regularly put cars through intense and unique stress tests. Vehicle components get dunked into nearly freezing vats of water and are frozen for weeks at a time in temperature and humidity-controlled chambers. Engineers monitor for failures and sensor health. Improvements and adjustments are made as necessary.
Can self-driving cars be hacked?
Engineers can take a number of steps to protect self-driving cars against cyberthreats and hackers. For example, Waymo uses layers of security to protect its self-driving system, especially safety-critical functions like steering and braking, against unauthorized communications, including vehicle control commands. Waymo’s vehicles do not rely on a constant wireless connection to operate safely. They have the ability to communicate with Waymo’s operations center to gather information about road conditions, while the vehicles themselves maintain responsibility for driving tasks at all times. All these communications are encrypted.
Waymo hs also implemented security mechanisms for detecting and analyzing anomalous behavior. Once detected, attempts to impair a vehicle’s security trigger an incident response procedure that involves impact assessment, containment, recovery, and remediation.
Learn more about how Waymo keeps its vehicles safe from cyberattacks.
Únase a nosotros en las conversaciones más importantes sobre cómo la tecnología de conducción autónoma puede dar forma al futuro de la seguridad, la movilidad, la comunidad y la sociedad.