Geometric Approach to Mapping and Localization

Features:

• Based on a abstract, compact representation of shape
• Utilizes state-of-the-art shape matching
• No odometry information required
• Reliable & Efficient
• Well-suited for object tracking and mapping applications

Principles of Operation:

1. Shape features are extracted from laser range finder (LRF) data
2. Shapes of two consecutive scans are matched against each other yielding a correspondence of objects
3. A map of robot environment composed of just a few polylines is built incrementally
   by merging polylines in a local scan with the existing global map

Demonstration videos:

Indoor environment at the University of Bremen:

View movie (measure 2) illustrating our incremental map building

View shape-based feature tracking
 

Simulated environment with noise added:

View movie illustrating our incremental map building

This project is supported by NSF under the grant entitled
Robot Localization and Robot Mapping Based on Shape Matching
as US-Germany Cooperative Research between Temple University and the University of Bremen, Germany
PI: Longin Jan Latecki

Collaborators:

Rolf Lakaemper (Temple University)

Diedrich Wolter and Christian Freksa (University of Bremen, Germany)

Xinyu Sun (Texas A&M University)