Master Thesis: Class-aware Pointcloud Filtering and Matching for Long-Term SLAM

Call for applications for the field of study, such as: Computer Science, Software Engineering, Mechanical Engineering, Mechatronics or similar  

 

In the research group Navigation Mobile Robots, we develop autonomous robots for a variety of outdoor applications, such as agriculture, forestry and logistics. The focus is on the development of an autonomous outdoor navigation solution as well as the hardware of the robots.

 

Be part of change

Since these outdoor applications have constantly changing environments due to weather conditions, seasonal variations, new constructions, and dynamic obstacles, the robot must continuously update its reference map of the whole environment. The localization module must understand which parts of the environment are static and which are semi-static or dynamic and update the global pointcloud reference map accordingly. Furthermore, the scan matching of the live pointclouds with this sparser global pointcloud map must be improved to get good pose estimate even with many dynamic obstacles blocking the view.

 

You will develop a global pointcloud mapping pipeline consisting of a pointcloud filtering approach and a pointcloud matching algorithm. The filtering approach should utilise camera-based classifications, from a sperate module, in the pointcloud data to distinguish different object types and construct and maintain a consistent global pointcloud map. The scan matching algorithm needs to be suitable for sparse environments with dynamic obstacles and must result in very few false positive matches. There already exists a basis of a 3D-LiDAR Localization module, that uses a local pointcloud map for odometry and a static global pointcloud map for absolute reference and your algorithm will replace the static global pointcloud map with your new updating global pointcloud mapping pipeline.
You will implement your solution inside our ROS2, C++ navigation stack and evaluate its performance both in simulation and in different real-world scenarios with our mobile CURT robots.