Visual Environment Assessment for Safe Autonomous Quadrotor Landing

TL;DR

This work addresses autonomous, GPS-denied quadrotor landing by fusing semantic segmentation with disparity-derived geometry to directly produce a 2D binary map of safe/unsafe landing zones, bypassing costly elevation maps. A cost-based site selection framework, J = $\alpha J_d + \beta \frac{1}{J_{un}}$ with $\alpha+\beta=1$, balances proximity to the drone and distance from hazards, guiding real-time landing decisions onboard a Jetson NX. Key contributions include a fully onboard, elevation-map-free pipeline that integrates metric and semantic cues, and a minimum snap trajectory for autonomous landing, demonstrated across multiple challenging indoor scenarios with high landing success rates. The approach offers practical impact for safe recovery and autonomous operation of aerial robots in GPS-denied or dynamic environments, especially under SWaP constraints.

Abstract

Autonomous identification and evaluation of safe landing zones are of paramount importance for ensuring the safety and effectiveness of aerial robots in the event of system failures, low battery, or the successful completion of specific tasks. In this paper, we present a novel approach for detection and assessment of potential landing sites for safe quadrotor landing. Our solution efficiently integrates 2D and 3D environmental information, eliminating the need for external aids such as GPS and computationally intensive elevation maps. The proposed pipeline combines semantic data derived from a Neural Network (NN), to extract environmental features, with geometric data obtained from a disparity map, to extract critical geometric attributes such as slope, flatness, and roughness. We define several cost metrics based on these attributes to evaluate safety, stability, and suitability of regions in the environments and identify the most suitable landing area. Our approach runs in real-time on quadrotors equipped with limited computational capabilities. Experimental results conducted in diverse environments demonstrate that the proposed method can effectively assess and identify suitable landing areas, enabling the safe and autonomous landing of a quadrotor.

Figures (5)

• Figure 1: Top: our drone navigating and mapping the environment. Bottom: the associated 2D binary map of the safe and unsafe landing locations (left) and the chosen safe landing spot in the 2D map (right). The black areas are the safe landing locations, while the gray ones are unsafe.
• Figure 2: Overview of our autonomous safe site detection and landing system: we use our quadrotor with a NVIDIA Jetson NX for computation and a stereo camera for VIO & mapping the environment. All our algorithms run in real-time onboard.
• Figure 3: Segmentation results in three different scenarios: on the left column the RGB images, on the right column the segmentation results. The green areas are considered unsafe.
• Figure 4: (a) Data acquisition & processing pipeline for the map creation and (b) Site evaluation and safe autonomous landing experiment in a low height, middle density environment scenario with an "8" navigation pattern.
• Figure 5: 2D binary map of safe and unsafe landing locations, overlayed to the real environment. The light green regions are unsafe while the dark green are unknown and still to be explored. Both of them are hazardous areas for landing.