Recursive Particle Filter with Geometric Constraints for SLAM

The multi-sensor fusion for a stochastic approach in the simultaneous localization and mapping (SLAM) should take into consideration the filtering performance against modeling error and required computational time. We propose a recursive particle filter based on the recursively marginalized Bayesian SLAM posterior considering geometric constraints. The recursively marginalized Bayesian SLAM posterior is based on approximating a traditional Bayesian SLAM as a product of the marginal probabilities related to estimates of the state and incorporates concurrently geometric constraints which keep geometric relationships between landmarks similar to correlations among landmarks. The proposed recursive particle filter based SLAM with geometric constraints allows a particle filter not only to directly update the state but also a larger number of particles to be incorporated for updating those particles may help the filtering performance. Furthermore, it can deal with any geometric constraints representing geometric relationships between landmarks which play an important role for stable map such as correlations among landmarks. The simulation result demonstrates a validity of the proposed method in estimation accuracy and efficiency