Autonomous Exploration and Mapping with RFS Occupancy-Grid SLAM.

Area of Science:

• Robotics
• Artificial Intelligence
• Sensor Fusion

Background:

• Autonomous mobile robots require efficient strategies for simultaneous localization and mapping (SLAM).
• Path-planning algorithms must balance exploration (map building) and exploitation (robot localization).
• Existing SLAM methods often struggle with joint map-pose uncertainty and require manual tuning.

Purpose of the Study:

• To develop an autonomous on-line path-planning algorithm for mobile robot exploration and occupancy-grid mapping.
• To propose a novel reward function for path-planning that integrates map and pose uncertainty.
• To improve the efficiency and robustness of SLAM systems.

Main Methods:

• Utilizing random-finite set (RFS) modeling for ranging sensor measurements within a Rao-Blackwellised particle filter framework.
• Implementing a path-planning strategy based on the Rényi divergence between prior and posterior probability densities.
• Developing a joint map-pose uncertainty measure derived from RFS modeling.

Main Results:

• The proposed reward function effectively balances exploration and exploitation without manual weight tuning.
• The algorithm provides a unified measure of map and pose uncertainty.
• Demonstrated improved performance in autonomous exploration and occupancy-grid mapping tasks.

Conclusions:

• The novel reward function based on Rényi divergence offers a principled approach to path-planning in SLAM.
• This method enhances autonomous robot navigation by optimizing the trade-off between exploration and localization.
• The RFS-based approach provides a robust solution for joint map-pose uncertainty management in mobile robotics.