H.A.W.K.: Hybrid Adaptive Waypoint Knowledge for Multi-Domain UAV Navigation

Authors: Sasiram Anupoju, Abhinav Samala, Rishikesh Tingirikar, Tharun Nalamasu

Abstract:
An AI-powered Unmanned Aerial Vehicle (UAV) navigation system is designed to understand natural
language instructions and interpret visual surroundings, enabling it to autonomously reach target
locations known as waypoints. Unlike traditional drone navigation systems that operate only in fixed or predefined environments, H.A.W.K. (Hybrid Adaptive Waypoint Knowledge) introduces a task- driven and adaptive framework that functions effectively across multiple unseen domains such as urban, coastal, and rural environments without requiring retraining. The system integrates computer vision for real-time scene understanding, natural language processing for interpreting user commands, and memory-based learning techniques to continuously improve navigation decisions during execution. Additionally, a lightweight domain adaptation mechanism allows the UAV to generalize its navigation strategies when transitioning between different environments, ensuring robust and flexible performance. With potential applications in aerial surveillance, disaster response, logistics, and traffic monitoring, this work contributes toward the development of intelligent, autonomous, and adaptable UAV navigation systems.

Keywords: Unmanned aerial Vehicle (UAV) Navigation, Hybrid Adaptive Waypoint Knowledge (H.A.W.K), Vision-Language Navigation (VLN), Domain Adaptation, Autonomous Systems, AirSim, Multi-Domain Learning.