We’ve recently had four papers accepted for presentation in the upcoming IROS 2017 and ITSC 2017 conferences. The topics of these works relate to improvements to the scan matching algorithms used by robots to build maps and localise, a novel way for robots to recognise what kind of place they are in, and more agile manoeuvering.
Zaganadis et al. have developed a scan matching algorithm that partitions 3D scans from the environment into flat and non-flat regions and then matches the regions in combination, and demonstrate that doing so improves the performance of 3D scan matching in certain scenarios.
Andreasson et al. have developed a scan matching algorithm that incorporates information from the robot’s wheel sensors in a novel way, and demonstrate improved scan matching performance particularly in corridors and aisles.
Magnusson et al. have shown a simple way of addressing the place categorisation problem, without the need for substantial manual training and labelling that has been required previously. This will enable robots to recognise what kind of place they are in, based on what it looks like (a kitchen vs an office, or a forest road vs a smaller track), and act accordingly.
The work of Banzhaf et al. will lead to better motion planning for tight manoeuvres, which may make it easier for warehouse robots to pick up goods in cluttered environments.