The Lab

Velodyne VLP-32C lidar

The long-range lidar helps to detect far-away obstacles, e.g., cars and pedestrians. The Velodyne VLP-32C has 32 lidar beams and has a detection range of about 80 meters. Ground removal and clustering is applied to extract obstacles from the point cloud. Alternatively, a special neural network can be used to detect other vehicles and pedestrians.

Ouster OS1-128 lidar

The short-range lidar helps to detect close-by and small obstacles around the vehicle. The Ouster OS1-128 produces a very dense point cloud with its 128 rotating laser beams and has detection range of around 40 meters. The precision of the lidar measurements is about 2–3 cm. Ground removal and clustering is applied to extract obstacles from the point cloud.

Allied Vision Mako G-319C cameras

The two cameras keep an eye on traffic lights. The left camera is pointed forward to see the farther traffic lights and the right one is pointed slightly to the right to see the closest traffic lights next to stop lines. The locations of the traffic lights are known from the high-definition map and are projected to the camera image. A patch with the traffic light is then extracted from the camera image and a neural network is used to classify whether the light is red, green or yellow.

Aptiv ESR 2.5 radar

The radar uses reflections of radio waves to estimate the distance to the vehicles ahead. It also produces instant speed measurements of those vehicles by utilizing the Doppler effect. The radar is very good at detecting large metallic objects, such as cars but struggles with pedestrians and smaller objects. Additionally, its distance measurements are very accurate in the longitudinal direction but lack precision in the lateral direction.

NovAtel PwrPak7D GNSS device

The GNSS device determines the location of the car using satellite positioning and corrections from nearby base stations. The system makes use of a Real-Time Kinematic (RTK) system with the ESTPOS base station network to consistently achieve a positioning accuracy of 5–10 cm. The car needs to know its precise position on the high-definition map to follow the virtual "rails" featured on the map.

AStuff Spectra computer

The computer runs our in-house software that makes all the driving decisions. Other than that, the computer is just a normal PC running an Ubuntu operating system.