Robots navigating through unknown terrain have to determine their own position while simultaneously creating a map of the environment around them. This is a reoccurring problem known as SLAM (Simultaneous localization and mapping) and is considered as one of the key problems to be solved on the road towards fully autonomous robots. Over the years various different solutions for this problem have been implemented successfully, although always for a specific platform equipped with a specific sensor set. This has resulted in monolithic pieces of software with limited reusability due to platform, algorithm and sensor specific dependencies and optimizations, making the effort of implementing SLAM on a new platform often unnecessarily complex and large.
This work aims to create and evaluate a component based SLAM framework that focusses on the design of well-defined interfaces and that allows implementation of various SLAM-algorithms on various platforms using either probabilistic or graph-based back-ends with minimal changes. In addition, if time allows, the goal is to implement active SLAM, which adds the element of active path planning in order to create the map efficiently and find unknown objects. In order to demonstrate the proof of concept, the framework is used to implement active SLAM on at least two different robot platforms: Loomo at Demean and the Husky at RaM.
Towards a modular sensor-independent active SLAM framework
Finished: 2020-05-26
MSc assignment