This research introduces Hier-SLAM++, a neuro-symbolic semantic mapping system that efficiently creates accurate 3D maps with semantic understanding while optimizing computational resources.

• Combines RGB-D and monocular input with 3D Gaussian Splatting for precise pose estimation
• Employs a novel hierarchical categorical representation to reduce parameter usage in complex environments
• Enables real-time semantic scene understanding with optimized computational efficiency
• Achieves accurate global 3D semantic mapping with streamlined processing

For engineering applications, this approach represents a significant advancement in spatial understanding for autonomous systems, robotics, and computer vision where resource efficiency is crucial.

Hier-SLAM++: Neuro-Symbolic Semantic SLAM with a Hierarchically Categorical Gaussian Splatting