Localization in Self Driving Cars
Localization is how an self driving car knows where it is in the world. In day to day life we use phone GPS to localize our self all the time, but GPS is not accurate enough for autonomous vehicles. most of the GPS is accurate in between 1 to 3 meters which is for imprecise for self driving cars. There are some cases like if we surrounding by mountains or tall buildings the the GPS precision might be even worse it’s in between 10 to 50 meters.
So we need to find other way to more precisely locate our vehicle on the map. The method used in self driving cars is to compare what sensors see to the what appears on the map. Vehicle sensors can measure the distance between the vehicle and the static objects such as trees, poles, road signs, walls. we measure the these distances and direction to theses static objects in the vehicle co-ordinate frame. The landmark detected by the vehicle sensors may also be on the map. To estimate the position of the vehicle on the map, we match the landmark observations of our sensors with the locations of those landmarks on the map. The vehicle has it’s own co-ordinates and the map also have its own co-ordinate system. Our self driving vehicle software has to transform the sensor measurement from the vehicle co-ordinate frame to the map co-ordinate system and vice-versa. Then the system must be pinpoint the vehicle precise location on the map to within 10cm.
We also use different methods to get the precise location of our vehicle. They are
- GNSS RTK
- Inertial Navigation
- Lidar Localization
- Visual Localization
GNSS RTK
Global navigation satellite system Real-time kinematic positioning is a satellite navigation technique used to enhance the precision of position data derived from satellite-based positioning systems (global navigation satellite systems, GNSS) such as GPS, BeiDou, GLONASS, Galileo and NavIC. It uses measurements of the phase of the signal’s carrier wave in addition to the information content of the signal and relies on a single reference station or interpolated virtual station to provide real-time corrections, providing up to centimetre-level accuracy.
Inertial Navigation
An inertial navigation system (INS) is a navigation device that uses a computer, motion sensors (accelerometers) and rotation sensors (gyroscopes) to continuously calculate by dead reckoning the position, the orientation, and the velocity (direction and speed of movement) of a moving object without the need for external references. Often the inertial sensors are supplemented by a barometric altimeter and sometimes by magnetic sensors (magnetometers) and/or speed measuring devices.
An Inertial Navigation System (INS) provides real-time roll, pitch, and heading integrated with a GNSS receiver to maintain accuracy in case of signal outages (buildings, trees, tunnel, etc. The inertial sensor is also used to precisely synchronize and stabilize additional equipment such as LiDAR or Camera for a driverless car application.
Lidar Localization
With lidar we can localize a car by means of point cloud matching. This method continuously matches the detected data from the lidar sensors with the pre-existing high definition map. This comparison yields the global position and heading of the car on the high-definition map. There are many algorithms to use for matching point clouds. Iterative closest point or ICP is one approach. Kalman filters are another approach to the lidar localization. A Kalman filter is an algorithm used to predict our current state based on our past state in new sensors measurements. Localization using LIDAR has advantages over visual localization. In particular LIDAR has a great degree of viewpoint and lighting invariance. It is however less informative lidar localization is robustness. Constructing and keeping up to date of high definition maps is the most complex task in fact never possible.
Visual Localization
Cameras images are easiest type of data to collect. Cameras are cheap and plentiful, easy to use. Precise localization through images is hard. In practice camera images are often combined with data from other sensors to accurately localize the vehicle. But camera data combined with map and GPS data can localize much better than camera images alone.
Here is is the link of my article on self driving cars intro:
We will discuss further more on self driving cars on next blog’s.