Context-Aware Semantic Localization with Adaptive Sensor Fusion Under Adverse Conditions

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Autonomous driving requires robust vehicle localization under challenging real-world conditions.
• Conventional methods struggle with sensor degradation and ambiguous data, leading to inaccurate pose estimates.

Purpose of the Study:

• To develop a semantic localization framework integrating ontology-based reasoning for enhanced accuracy and reliability.
• To address limitations of geometric-based localization in degraded sensor conditions.

Main Methods:

• Reformulated localization as a context-aware constraint selection problem.
• Integrated semantic reasoning to evaluate logical and contextual validity of pose hypotheses.
• Employed ontology-based semantic consistency checks for objects, places, and vehicle poses.

Main Results:

• Achieved an average reduction of 35.6% in mean localization error and 47.0% in maximum localization error.
• Demonstrated improved robustness and accuracy, particularly under sensor degradation and dynamic environments.
• Enabled structured multi-sensor fusion by selecting semantically relevant information.

Conclusions:

• Semantic localization significantly enhances autonomous vehicle positioning accuracy and reliability.
• The framework reduces computational complexity and is adaptable to cooperative perception systems.
• Ontology-based semantic reasoning is crucial for next-generation autonomous driving localization.