Log in

Quick registration

What does a self-driving car see as the world around it?

Author:Constantine Time:2018/03/31 Read: 5919
When a self-driving car looks at the world, it sees a lot. It has radar to measure the distance to the car next to it, and it has cameras to capture color images of the street […]

When a self-driving car looks at the world, it sees a lot. It has radar that measures the distance to the car next to it, it has cameras that capture color images of the street, and its lidar sensor emits laser pulses to measure its surroundings. For any robot-driven car, one of the most important parts of the ride is not what it sees, but what it knows in advance about the road it's passing.

A woman died after being hit by an Uber self-driving car in Tempe, Arizona, on the evening of March 18, according to a statement from local police. The car was operating on autonomous driving at the time of the incident, although there was a safety driver behind it. Uber is one of many companies now testing such vehicles in Arizona, California and other parts of the country. Waymo, the self-driving car company made by Google parent Alphabet, said it also operates self-driving cars in suburban Phoenix and does not have safety drivers. On March 19, Uber said it was conducting testing in Tempe, Pittsburgh, Toronto and San Francisco.

When designing these vehicles, companies like Uber and Waymo first build a three-dimensional map of a place. They equipped regular cars with lidar sensors, "light detection and ranging" devices that use pulses of light to measure distance. As company workers drive the cars on local roads, the expensive equipment collects the information needed to create the maps. Once the map is complete, the car can use it to navigate the road on its own. As they do so, they continue to use lidar to track their surroundings and compare what they see to what the map shows. This way the car gets a good idea of its place in the world.

Expand the rest58%

Lidar also warns the car about nearby objects, including other cars, pedestrians and cyclists. The radar works great, but it can't do anything. It only provides information about relatively close objects, which limits how fast the car can drive. Its measurements are not always clear enough to distinguish each object. Moreover, when multiple self-driving vehicles travel on the same road, their lidar signals may interfere with each other. Even when lidar is working well, these companies need backup systems.

So most self-driving cars are also equipped with a variety of other sensors. With cameras, radar, and GPS antennas, this GPS hardware tells your smartphone where it is. With GPS antennas, companies like Uber and Waymo are giving cars more information about their location. Using cameras and radar sensors, they can collect additional information about nearby pedestrians, cyclists, cars and other objects. And the cameras also provide a way to identify traffic lights, street signs, road markings and other signals that cars need to take into account.

In some cases, engineers write specific rules that define how the car should respond in certain situations. For example, a Waymo car is programmed to stop if it detects a red light. But engineers may not be able to plan responses for every situation a car encounters. So companies like Waymo and Uber have come to rely on "machine learning" systems to learn behavior by analyzing vast amounts of data describing the country's roads. Waymo is using a system to identify pedestrians by analyzing thousands of photos of people walking, running across or near roadways. It's unclear what happened to Uber's self-driving cars in Tempe. But these cars are designed so that if one system fails, another system will kick in.

Self-driving cars may struggle to replicate the subtle nonverbal communication between pedestrians and drivers. After all, self-driving cars can't make eye contact with people in crosswalks. Self-driving cars are still struggling to master these conditions, though. But they don't work well in heavy precipitation, they can have trouble in tunnels and bridges, and they also have trouble facing heavy traffic. Therefore, whether it is considered from the perspective of technological maturity, production cost or safety supervision, driverless cars are still a long way from becoming mainstream cars. (New technological discoveries by Constantine/Text)



Leave a Reply


copyright © www.scitycase.com all rights reserve.
Beijing ICP No. 16019547-5