Particle Filter Networks with Application to Visual Localization
Particle filtering is a powerful method for sequential state estimation and is extensively used in many domains, including robot localization, visual tracking, etc. To apply particle filters in practice, a main challenge is to construct an effective probabilistic system model, especially when the system exhibits complex dynamic behavior or processes rich sensor information from, e.g., visual cameras. This paper introduces the Particle Filter Network (PF-Net), which captures both a system model and the particle filter algorithm in a single neural network. This unified network representation enables end-to-end model learning, which trains the model in the context of a specific algorithm, resulting in improved performance, compared with conventional model-learning methods. We apply PF-net to visual robot localization. The robot must localize in rich 3-D environments, using only a schematic 2-D floor map. In preliminary experiments, PF-Net consistently outperformed alternative learning architectures, as well as conventional model-based localization methods. PF-net learns effective models that generalize to new, unseen environments. It can also incorporate semantic labels on the floor map.
READ FULL TEXT