AirNav:A Computation-driven Cloud-based Solution for the Drone Path Planning Utilizing Machine Learning

摘要

AirNav represents a significant advancement in drone technology, offering a cloud-based path planning system that effectively addresses the critical limitations of UAVs. By leveraging cloud computing and artificial intelligence, AirNav enhances UAV capa- bilities and operational efficiency, enabling seamless communication and control over the Internet and effectively eliminating dis- tance restrictions in operations. The system’s hybrid architecture dynamically offloads computationally intensive tasks, including reinforcement learning-based path planning, to cloud resources. This innovative approach not only reduces hardware complexity and overall costs but also significantly improves navigation efficiency and extends operational time. AirNav incorporates adaptive energy and delay-sensitive offloading strategies, ensuring optimal resource utilization and enhanced performance across diverse environments. Implemented using AWS services such as SageMaker, Lambda, and S3, AirNav provides a scalable and flexible platform for advanced UAV management. The system employs a novel threshold-based dynamic computation offloading strategy, balancing onboard and cloud processing to optimize energy consumption and response times. This approach has demonstrated remarkable improvements, with adaptive thresholding outperforming traditional methods by 60% in response time and requiring 16.67% less data transfer. Furthermore, the cloud-based SageMaker Instance shows an average reduction in energy consumption of about 37.5% compared to local onboard systems. By virtualizing access through Web services and supporting HTTP/HTTPS over RESTful APIs, AirNav establishes a robust framework for next-generation drone operations. This comprehensive solution not only addresses current challenges in UAV technology but also paves the way for more efficient and capable drone applications across various industries, marking a significant step forward in the evolution of aerial robotics.