Speaker
Description
Unmanned Aerial Vehicles (UAVs) are critical tools in applications ranging from disaster response to infrastructure inspection. However, their operational efficacy is often hindered by an inability to land safely on uneven or unpredictable terrain, limiting utility in dynamic environments. This research presents an AI-enhanced adaptive landing system that transforms conventional UAVs into versatile aerial-ground hybrid platforms capable of operating in dynamic environments. The system integrates a hierarchical reinforcement learning (HRL) framework with multimodal sensor fusion to optimize landing strategies and terrain adaptation. A novel hexapod landing gear design—featuring telescopic legs with articulated joints—allows rapid (<2 seconds) reconfiguration between aerial and terrestrial locomotion. A deep learning-based terrain classification framework dynamically adjusts leg kinematics during descent, while compliant mechanisms minimize impact forces. Key innovations include sensor-guided stability optimization and a hardware-software co-design that reduces collision risks during inspection tasks. Experimental validation through simulation prototyping demonstrated a 52% improvement in post-landing stability on irregular surfaces, a 68% reduction in landing impact forces, and 45% greater positional precision compared to traditional fixed landing gear. These advancements significantly enhance mission reliability in scenarios such as structural inspections, precision agriculture, and disaster zones where unstable terrain is prevalent. By addressing critical gaps in UAV-ground interaction, this work advances the autonomy of hybrid aerial-ground robots, enabling safer and more versatile operations in unstructured environments. The system’s modular design also offers broader implications for adaptive robotics in dynamic real-world applications.
