Nvidia, an American technology company based in Santa Clara, California released a video at the CES 2016 showing the modeling environment to be used for self driving cars. The impressive model shows how a standard car was equipped with 6 1-megapixel cameras and 4 lidars. The 2 lidars built in the front and rear of the car are for collision prevention, plus 2 lidars on the roof focusing on the reconstruction of the surroundings.
The team stated that there are many configurations that can be used to determine the model environment. However, they chose this particular configuration to test out the entire platform of self driving cars.
A team member was sitting at the back of the car while the video was streaming from the car cameras directly into a computer. See the car in action in the video below.
What information does the car receive?
The first goal was to determine where the car is by using sensors. This will help measure the car's motion. 4 fish eye cameras were taking 8,000 points per camera (30 frames per second) to create a reconstruction that puts the car in 3D space. At the same time, lidar data provided the distance of points where the lidar beams hits through space.
With all of these data from the cameras and lidars, an Occupancy Map was created. The map shows spaces around the car that is not being occupied by another object. As vehicles move past the car, the white points around the car fade away. This gives the car information that the other lanes are available or occupied.
The next step is to lay out where the car is on the map.
Different algorithms were written for path planning, map planning, and navigation. Based from the experiment that they conducted on the Santa Clara freeway, GPS is simply not enough for self driving cars because it is only accurate in meters. For safety reasons, accuracy is important down to the last centimeter (obviously).
The video also shows that the team has created a simplified path finder. Using the cameras and sensors, the car is being fed with information if the lanes around it are occupied or not. With allowable measurements, the car can safely change lanes without colliding with other vehicles.
This video was presented at the recent CES 2016 held in Las Vegas from January 6 to 9, 2016. And for an added bonus, see Chris Urmson cover how a driverless car sees the road: