EDGE AI FOR AUTONOMOUS SYSTEMS: REAL-TIME DEEP LEARNING ALGORITHMS FOR SENSOR FUSION AND DECISION-MAKING IN DRONE NAVIGATION

Abstract

This study explores the integration of real-time deep learning algorithms and sensor fusion techniques for autonomous drone navigation, focusing on the use of edge AI to optimize system performance under constrained computational resources. The research demonstrates the critical role of multi-sensor fusion—incorporating camera, LiDAR, IMU, and GPS—in improving the accuracy and reliability of drone navigation systems. The hybrid fusion model achieved an impressive 95% accuracy in environmental perception, significantly outperforming individual sensor modalities. Convolutional neural networks (CNNs) in object identification applications possess 50 ms of inference time enabling real-time operation.  The reinforcement learning (RL) models exhibited superior decision-making capabilities through their slower operation speed (150 ms). The hybrid RL model cut down 100-meter route travel time to achieve 25 seconds which optimized overall route planning effectiveness.  According to measurements the CNN model operated with 3.5 W as the most economical solution while the hybrid approach provided reasonable power-management compared to performance output.  A sensor fusion model decreased its response time to 90 milliseconds which secured quick reactions needed for changing environments.  Edge artificial intelligence demonstrates its ability to enhance autonomous drones through better navigation precision and decision capabilities and efficiency which makes them suitable for real-time energy-efficient applications such as environmental observation and search and rescue operations.  Results demonstrate that autonomous navigation requires both deep learning and sensor fusion to build a stronger dependable autonomous navigation system.