Sampling Strategies for Particle Filtering SLAM Kristopher R. Beevers We describe several new sampling strategies for Rao-Blackwellized particle filtering SLAM. Two of the strategies, called fixed-lag roughening and the block proposal distribution, attempt to exploit “future” information, when it becomes available, to improve the filter’s estimation for previous time steps. Fixed-lag roughening perturbs trajectory samples over a fixed lag time according to a Markov Chain Monte Carlo kernel. The block proposal distribution directly samples poses over a fixed lag from their fully joint distribution conditioned on all the available data. Our results indicate that the proposed strategies, especially the block proposal, yield significant improvements in filter consistency and a reduction in particle degeneracies compared to standard sampling techniques such as the improved proposal distribution of FastSLAM 2. In addition, we examine the effectiveness of two new resampling techniques, residual resampling and generalized resampling, as applied to RBPF SLAM. These drop-in-place techniques are simple to use and (in the case of residual resampling) computationally cheaper than the standard random resampling approach. However, our results show that they offer no real improvement in performance over random resampling in SLAM. This is an extended version of a paper (Beevers and Huang, 2007) previously submitted for publication. Department of Computer Science, Rensselaer Polytechnic Institute, Troy, NY cs-06-11
Sampling Strategies for Particle Filtering SLAM
Kristopher R. Beevers
We describe several new sampling strategies for Rao-Blackwellized particle filtering SLAM. Two of the strategies, called fixed-lag roughening and the block proposal distribution, attempt to exploit “future” information, when it becomes available, to improve the filter’s estimation for previous time steps. Fixed-lag roughening perturbs trajectory samples over a fixed lag time according to a Markov Chain Monte Carlo kernel. The block proposal distribution directly samples poses over a fixed lag from their fully joint distribution conditioned on all the available data. Our results indicate that the proposed strategies, especially the block proposal, yield significant improvements in filter consistency and a reduction in particle degeneracies compared to standard sampling techniques such as the improved proposal distribution of FastSLAM 2. In addition, we examine the effectiveness of two new resampling techniques, residual resampling and generalized resampling, as applied to RBPF SLAM. These drop-in-place techniques are simple to use and (in the case of residual resampling) computationally cheaper than the standard random resampling approach. However, our results show that they offer no real improvement in performance over random resampling in SLAM. This is an extended version of a paper (Beevers and Huang, 2007) previously submitted for publication.
Department of Computer Science, Rensselaer Polytechnic Institute, Troy, NY
cs-06-11