Our primary research goal is to develop novel and effective methods to enable effective robotic mapping and localization, allowing a autonomous mobile robots to accomplish safe and efficient navigation. Our research interests include robot vision, pattern recognition, machine learning, scene understanding, visual tracking, filtering & optimization, database & information retrieval, and data compression & recovery.
Approach: Map Matching
Since 1995, we are particularly interested in map matching, the ability to match a local map built by a mobile robot to previously built maps, which is crucial in many robotic mapping, self-localization, and simultaneous localization and mapping (SLAM) applications. The goal of map matching is given a pair of input maps, to find a transformation (e.g., rotation, translation) from one map to the other. One of most poularsolusions to the map matching is RANSAC, in which a number of hypotheses of the transformation are genereated from a minimal set of map features while each hypothesis is verified using the remained features as a cue. The other solutions include information reduction with dimension reduction techniques. The map matching task is particulary important in the following contexts:
Map updation. Change detection is an important task for updating maps in non-stationary environments. Map matching between a pair of environment maps generated at two different points in time provide cues for change detection.
Hongbin Zha, Kanji Tanaka, Tsutomu Hasegawa, Detecting Changes in a Dynamic Environment for Updating its Maps by Using a Mobile Robot, Proc. IEEE/RSJ Int. Conf. Intelligent Robots and Systems (IROS), 1997.
Map merging. Map matching is also important tool for aligning multiple maps with different coverages and merging them into a global environment map.
Kanji Tanaka, Eiji Kondo, Incremental RANSAC for online relocation in large dynamic environments, Proc. Int. Conf. Robotics and Automation (ICRA), 2006
Map retrieval. Map matching is a standard building block of map retrieval. Map retrieval is usually an important first step of global robot self-localization with respect to an environment map.
Kanji Tanaka, Tsutomu Hasegawa, HongbinZha, Eiji Kondo, Nobuhiro Okada, Mobile Robot Localization with an Incomplete Map in Non-Stationary Environments, Proc. Int. Conf. Robotics and Automation (ICRA), 2003
Map co-segmentation. Map co-segmentation, the task of discovering map segments that common appear in multiple different maps, is important for unsupervised map compression and map segmentation. Map matching is an essential building block of map co-segmentation.
Tomomi Nagasaka, Kanji Tanaka, Dictionary-based Map Compression for Sparse Feature Maps, Proc. Int. Conf. Robotics and Automation (ICRA), 2011