Deep Samplable Observation Model for Global Localization and Kidnapping

Global localization and kidnapping are two challenging problems in robot localization. The popular method, Monte Carlo Localization (MCL) addresses the problem by iteratively updating a set of particles with “sampling-weighting” loop. Sampling is decisive to performance MCL [1]. However, traditional can only sample from uniform distribution over state space. Although variants propose different sampling models, they fail provide an accurate or generalize across scenes. To better deal these problems, we present proposal model named Deep Samplable Observation Model (DSOM). DSOM takes map 2D laser scan as inputs outputs conditional multimodal probability pose, making samples more focusing on regions higher likelihood. With such samples, convergence expected be effective efficient. Considering that learning-based may capture pose sometimes, furthermore Adaptive Mixture (AdaM MCL), which deploys trusty mechanism adaptively select mode for each particle tolerate this situation. Equipped DSOM, AdaM achieve estimation, faster scalability than previous methods both synthetic real Even environments long-term changes, able localize using trained simulation observations SLAM blueprint map. Source code paper available here: https://github.com/Runjian-Chen/AdaM_MCL.