Context-Aware Fail-Safe Multi-Modal Sensor Fusion Framework for Autonomous Vehicle Navigation in an Indoor Environment

Abstract

Indoor autonomous vehicles face persistent localization challenges due to sensor failures, path restrictions, illumination changes, and dynamic obstacles.We propose a fail-safe, context-aware, multi-modal sensor fusion framework to ensure reliable navigation. The approach combines ensemble localization, confidence-weighted fusion via Kalman consensus, and a fail-safe decision algorithm integrated with adaptive path planning. Sensor weights are dynamically adjusted to maintain accuracy under partial or complete modality failures. We used the ”Haram” (pilgrimage ritual (Tawaf)) as an inspirational application for the simulation and evaluation of the proposed framework. For this purpose,we created an AV cart and replicated it by maintaining the challenges of an interior setting, a comparable path length, and occlusions. The work covers the equivalent environmental applicability to other complex environments such asairports, hospitals, and shopping malls. The framework is implemented in ROSwith a 3D-customized cart navigating circular and elliptical paths under simulated sensor networks. Performance evaluation using Average Displacement Error DE), Final Displacement Error (FDE), collisions, and trip completion time shows improved robustness and higher success rates than static fusion, even under
repeated sensor failure scenarios.