Feature-Based Occupancy Map-Merging for Collaborative SLAM.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Probabilistic occupancy grid maps are crucial for collaborative mapping in robotics.
• Exchanging and integrating maps among robots reduces exploration time, a key benefit of collaborative systems.
• Map fusion necessitates solving the challenging unknown initial correspondence problem.

Purpose of the Study:

• To present an effective feature-based map fusion approach for collaborative mapping.
• To address the unknown initial correspondence problem in merging probabilistic occupancy grid maps.
• To enable robust and accurate global map construction from multiple individual maps.

Main Methods:

• Feature-based map fusion processing spatial occupancy probabilities.
• Detection of features using locally adaptive nonlinear diffusion filtering.
• Verification procedure to ensure correct transformation and avoid ambiguous map merging.
• Global grid fusion strategy employing Bayesian inference, independent of merging order.
• Hierarchical map fusion for merging multiple maps simultaneously.

Main Results:

• The presented method successfully identifies geometrically consistent features across diverse mapping conditions.
• Effective handling of low overlapping areas and varying grid resolutions during map fusion.
• Demonstrated suitability for constructing a consistent global map for SLAM from six individual maps.

Conclusions:

• The proposed feature-based map fusion approach provides an effective solution for collaborative mapping.
• The method ensures robust and accurate integration of probabilistic occupancy grid maps.
• It significantly contributes to improving the efficiency and consistency of SLAM systems in collaborative environments.