A FootSLAM Gallery - For example our new buildung:
A FootSLAM Gallery can be found here.
FootSLAM/PocketSLAM uses inertial-based measurements such as pedestrian dead reckoning or NavShoe step measurements as the basis for computing the underlying building structure. We have performed experiments where a person wearing a foot mounted IMU walked in our office environment for roughly 10-15 minutes. The data was pre-processed with a Kalman filter to obtain step estimates (see section III C in this paper) and then processed in a sequential Rao-Blackwellized Particle Filter (RBPF) in a typical Fast-SLAM factorization. It's important to point out that no visual or ranging sensors were used; FootSLAM's only features or landmarks are the probability distributions of human motion as a function of location. See the papers below for experimental results and derivation of the Bayesian filter and RBPF. More information ca be found in our publication list [Publications]
PlaceSLAM is an extension to odometry based SLAM for pedestrians that incorporates human-reported measurements of recognizable features, or "places" in an environment. PlaceSLAM uses a spatial representation of such places can be built up during the localization process. We see an important application to be in mapping of new areas by volunteering pedestrians themselves, in particular to improve the accuracy of "FootSLAM" which is based on human step estimation (odometry). We distinguish between two important cases which depend on whether the pedestrian is required to report a place's identifier or not. Results based on experimental data show that the approach can significantly improve the accuracy and stability of FootSLAM and this with very little additional complexity. After mapping has been performed, users of such improved FootSLAM maps need not report places themselves. [ELIB] [IEEE Xplore] FootSLAM is also extended to handle known locations. [ELIB] [IEEE Xplore]
FeetSLAM is simply cooperative FootSLAM. The objective is that data from many walks can be combined to generate a more accurate and more encompassing total FootSLAM map. We have implemented an iterative processing algorithm motivated by Turbo Decoding from channel coding theory that takes maps from one data set as prior maps for other data sets. In two experiments performed so far we show that the algorithm improves the mapping accuracy with increasing iterations. The results were published at ION GNSS 2011. See thesis, videos and slides below.
WiSLAM uses RSS (received signal strength) measurements from Wifi access points (APs) as additional measurements. In the case of using RSS, WiSLAM needs to estimate not only the location, but also the effective transmit power of each AP. We have adopted a RBPF approach in which every particle carries its own WiSLAM map. WiSLAM can be added (well, multiplied in the likelihood function ...) to FootSLAM/PlaceSLAM. We use a Gaussian Mixture Model to approximate the location of the APs and assume a descrete PDF of the AP's effective transmit power. Our work was presented at IPIN 2011 [IEEE Xplore].
FootSLAM may be combined with a Prior Map to improve the position accuracy and early convergence. We propose to use Angular PDFs generated from floor plans to represent prior map for FootSLAM. The FootSLAM approach supported by additional prior-maps is very flexible: The prior map may represent only the outer walls of a building, and perhaps some of the inner walls. Because FootSLAM can over time learn the correct map, it is inconsequential if a number of walls are missing or erroneous because the map will be corrected over time. If a map is available it is profitable to use it as prior map. More informations about angular PDFs used as maps in pedestrian navigation can be found on our mobility models webpage Mobility Models. This work was presented at PLANS 2012 [ELIB] [IEEE Xplore].
FootSLAM is extended for integrating moving platforms (elevators and escalators). Machine learning and other techniques are investigated for moving platform detection. A new proposal function is defined for FootSLAM for integrating moving platforms in FootSLAM. With this, the platform movement can be emulated so that the position differences are reproduced in the calculated trajectory. [Elib][Link]
Successive collaborative FeetSLAM is developed in order to enhance FootSLAM under no loop closure/revisiting areas conditions. This technique can be applied either to multiple users following similar paths or one user wearing multiple sensors. For reducing complexity, the trajectories are divided into small portions (sliding window technique) and are partly successively applied to the collaborative SLAM algorithm. The results show that the mean position error can be reduced to ~0.5m when applying partly successive collaborative SLAM. [Elib][Link]
More Information on FootSLAM:
Publications Slides Video