Multimodal Obstacle Detection and Adaptive Neural Control for Autonomous Drones

Abstract

Achieving reliable navigation for autonomous drones in complex environments remains a significant challenge, particularly in low-light conditions. To address this, we propose an integrated multimodal obstacle detection and adaptive neural control system with online learning to enable drones to navigate autonomously both during the day and at night. The proposed multimodal obstacle detection system integrates two ranging LiDAR sensors and a depth camera with sensory processing techniques, including the iKD-Tree interested area search algorithm, sensor fusion, and neuro-obstacle directional feature extraction. This ensures robust obstacle detection across various conditions without requiring sensor reconfiguration. The adaptive neural control system applies Hebbian correlation-based learning and synaptic scaling plasticity principles to continuously update the control weights, allowing the drone to dynamically adapt its speed and maneuver around obstacles in real time. We evaluate the system’s performance in both simulation and real-world environments, demonstrating its effectiveness under diverse lighting conditions and obstacle types.