Object-aware semantic mapping using probability density functions for indoor relocalization and path planning.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Indoor robots require scalable semantic maps for complex environments.
• Existing semantic maps are either too dense and computationally expensive or lack geometric detail.
• A trade-off exists between map detail and computational efficiency, limiting real-world applications.

Purpose of the Study:

• To develop a novel object-aware semantic mapping framework for indoor robots.
• To create a compact, robust, and scalable semantic representation that balances detail and efficiency.
• To enhance robot capabilities in relocalization, path planning, and scene understanding.

Main Methods:

• Modeling key static objects using probability density functions (PDFs).
• Detecting objects via 3D point cloud processing and encoding them as 2D probabilistic occupancy distributions.
• Utilizing Differential Evolution and Kullback-Leibler divergence for robust relocalization without prior pose.

Main Results:

• The proposed framework provides a compact and robust representation preserving semantic identity and geometric shape.
• It effectively handles noise and partial views, enabling global relocalization and semantically informed path planning.
• Demonstrated improved performance over traditional methods in ambiguous or cluttered scenes.

Conclusions:

• The object-centric, probabilistic mapping framework offers a unified representation for multiple robotic behaviors.
• This approach supports functional scene understanding for context-aware navigation.
• Validated on benchmark datasets and real-world apartment environments, showing significant advantages.