Continually improving large scale long term visual navigation of a vehicle in dynamic urban environments

This paper is about long term navigation in dynamic environments. In previous work we introduced a framework which stored distinct visual appearances of a workspace, known as experiences. These are used to improve localisation on future visits. In this work we introduce a new introspective process, executed between sorties, thats aims by careful discovery of the relationships between experiences, to further improve the performance of our system. We evaluate our new approach on 37km of stereo data captured over a three month period.