Our team develops a tight coupled GNSS/INS/LiDAR/HD Map-based localization framework to provide robust positioning for autonomous vehicles in Super-Urbanized areas. Conventional GNSS/INS/LiDAR/HD Map-based can provide robust solution in open space or sub-urban area by integrating GNSS, INS, LiDAR and HD Map. However, this solution can be severely degraded due to the tall buidings and moving objects, especially in the scenario with tall buildings and heavy traffic. Firstly, performance of low-cost GNSS receiver-based solution can be decreased to even 100 meters due to the NLOS receptions and multipath effects and its error covariance is unknown. Moreover, commonly used RTK GNSS in autonomous driving can also be suffered whose positioning error can also go up to 20 meters with unknow error covariance. Secondly, INS solution can be suffered from drift due to the long time traffic congestion. Thirdly, mapping between LiDAR point cloud and HD Map can also undergo failure as the filed of view (FOV) for LiDAR can severely blocked by moving objects on the road, for example the double-decker bus in Hong Kong and London.
Github code: https://github.com/weisongwen/Autoware
Thus, our team is studying the innovational method to solve the following problem:
Multipath effects introduced by tall buildings will be mitigated, and NLOS receptions caused by tall building will be excluded by integrating 3D city models. However, moving objects with high altitude, such as the double-decker bus, can also cause NLOS receptions in deep urban area and this positioning error source will be mitigated by detecting the moving objects using LiDAR sensor.
Illustration of Double-decker bus detection using Euclidean cluster algorithm and parameters-based classification. Blue box represents the initially detected double-decker bus.
Skyplot visualization for satellites and double-decker bus boundary. Green circles and the nearby numbers indicates satellites and corresponding PRNs. Line segment indicates the boundary
Precise positioning is one of the key technology in this autonomous driving development. It is relatively easy to obtain its precise position (i.e., 10 centimeters) using real time kinematic (RTK) in open sky area, such as highways. Considering the vehicle will be operated in all kinds of environment, the positioning system should be able to provide robust localization information in harsh environment. Most of the driving environment is blocked by trees or high buildings, it is especially difficult to get a “fixed” solution for RTK algorithm in such environments. To solve, the problem a position hypothesis method is developed. This research is still under development. More result will be released soon.
Illustration of double difference of GNSS measurements between rover and reference station.
Targeting test environments in Hong Kong
Similarity Comparison between Standalone Position Hypothesis (PH), PH aided by DGPS correction, PH aided by measurement from reference station.