[論文レビュー] Visual Marker Search for Autonomous Drone Landing in Diverse Urban Environments

Jiaohong Yao, Linfeng Liang+4 more|arXiv (Cornell University)|Jan 16, 2026

UAV Applications and Optimization被引用数 0

ひとこと要約

The paper builds a simulation-based evaluation suite in AirSim to study autonomous visual-marker search and landing across diverse urban scenes, comparing two heuristic search strategies and a learning-based agent.

ABSTRACT

Marker-based landing is widely used in drone delivery and return-to-base systems for its simplicity and reliability. However, most approaches assume idealized landing site visibility and sensor performance, limiting robustness in complex urban settings. We present a simulation-based evaluation suite on the AirSim platform with systematically varied urban layouts, lighting, and weather to replicate realistic operational diversity. Using onboard camera sensors (RGB for marker detection and depth for obstacle avoidance), we benchmark two heuristic coverage patterns and a reinforcement learning-based agent, analyzing how exploration strategy and scene complexity affect success rate, path efficiency, and robustness. Results underscore the need to evaluate marker-based autonomous landing under diverse, sensor-relevant conditions to guide the development of reliable aerial navigation systems.

研究の動機と目的

• Motivate robust marker-based landing in urban settings where GPS is unreliable.
• Create a diverse, interactive simulation dataset reflecting urban variability (layout, lighting, weather).
• Benchmark navigation strategies for autonomous marker search under partial observability and sensor limits.

提案手法

• Use Unreal Engine 4 / AirSim to generate 966 episodes across three urban maps with varied lighting and weather.
• Evaluate three navigation strategies: Spiral (2D/3D), Zigzag (2D/3D), and E2E-RL (learning-based).
• Rely on onboard RGB for marker detection and depth for obstacle avoidance, with a pretrained detector at evaluation.
• Compare strategies using metrics like success rate, navigation error, SPL, collision rate, and false detection.
• E2E-RL trained with PPO and curriculum learning in a surrogate environment; detector kept separate during learning to avoid bias.

実験結果

リサーチクエスチョン

• RQ1 How do different exploration strategies affect robustness and efficiency of marker search in diverse urban environments?
• RQ2 What is the impact of scene layout, lighting, and weather on marker detection and navigation performance?
• RQ3 Can a learning-based agent outperform heuristic searches in terms of safety and efficiency in realistic urban settings?
• RQ4 How does performance vary across ModernCity, PostSoviet, and UrbanDistrict environments?

主な発見

• Heuristic methods (Spiral, Zigzag) achieve higher SR in some maps but suffer from higher collision rates.
• 3D variants (Spiral-3D, Zigzag-3D) improve SR and reduce NE overall compared to their 2D counterparts.
• E2E-RL shows lower collision rate and more efficient trajectories (higher SPL) but often lower SR due to less exhaustive search.
• Environment type significantly influences results; UrbanDistrict favors coverage-based methods, while PostSoviet presents vertical obstacles that impede performance.
• False detections and navigation errors correlate with higher initial altitude and degraded visibility in some scenarios.
• Across maps, no single method dominates; performance depends on scene geometry, obstacle density, and lighting.