Mobile robot localisation with incremental PCA

Mobile robots can employ different sensors to collect data about the environment. We propose to use a special sensor (a catadioptric camera) which provides panoramic views at each robot position. To model the environment for later navigation and localisation, we build a representation of the appearance by compressing the set of panoramic images by principal component analysis (PCA). Since the batch application of the PCA is inappropriate in this case, we propose to apply an incremental approach. This leads to novel aspects regarding the adaptation of compressed partial representation. We provide empirical results which indicate the performance of the proposed method is comparable to the performance of the batch method in terms of compression, computational cost, and, most importantly, precision of localisation.