AI-Driven Landing Zone Detection Module for Vertical Take-Off and Landing Vehicles Using Projection-Based LiDAR-Navigation Pipelines
Nirasha Herath, Oscar De Silva, George K. I. Mann, Awantha Jayasiri
AI summary
Problem
VTOL vehicles require rapid and accurate detection of safe landing zones in unknown terrains, but existing AI methods struggle with low-resolution, non-repetitive aerial LiDAR data or lack the computational efficiency needed for real-time deployment.
Approach
The authors developed a projection-based point cloud semantic segmentation CNN that converts raw LiDAR data into range images, combined with point cloud accumulation and hardware acceleration to handle low-resolution aerial scans efficiently.
Key results
- >0.93 mIoU and >98% accuracy across three aerial datasets
- >10 Hz processing rate and >5 million pts/s throughput on Jetson AGX Xavier
- Generation of ~52 million labeled points for LZ/non-LZ semantic segmentation
- Successful real-time validation in a VTOL LiDAR-navigation pipeline
Why it matters
Provides a computationally efficient, sensor-agnostic solution that enables safe autonomous VTOL operations in emergency and remote missions without relying on high-resolution sensors or camera fusion.
Abstract
This paper introduces an artificial intelligence-based landing zone detection module (LZDM) for vertical take-off and landing (VTOL) navigation. It employs a projection-based point cloud semantic segmentation (PCSS) convolutional neural network model combined with point cloud accumulation and a range image generation module. The proposed method addresses the limitations of existing projection-based PCSS methods, which often struggle with low-resolution and non-repetitive scan raw light detection and ranging (LiDAR) data commonly found in aerial datasets. The proposed LZDM was developed using three sets of aerial datasets collected from a DJI M600 hexacopter drone, a DJI M300 RTK quadrotor, and a Bell412 helicopter. The results were evaluated using both qualitative and quantitative metrics, demonstrating its robustness and effectiveness. In terms of quantitative results, the proposed method achieved mean intersection over union and accuracy values greater than 0.93 and 98 percent, respectively, across all three datasets, highlighting its accuracy in identifying safe landing zones (LZs). To assess the real-time feasibility of the proposed LZDM, it was deployed on a reconfigurable hardware-accelerated module. This setup achieved processing rates higher than 10 Hz for all three datasets and a throughput of over 5 million pts/s on the Jetson AGX Xavier dedicated hardware combined with the PyTorch TensorRT optimization module. The supplementary materials, including the inference code, sample testing data, and instructions are available in our GitHub repository at https://github.com/ nira16herath/CENet-LZ-detection/tree/main Note to Practitioners—VTOLs, such as small drones and helicopters, are increasingly becoming popular in time-critical and safety-critical operations, including parcel delivery to remote areas, search and rescue, and other autonomous missions. While advanced trajectory planning systems are available to guide these vehicles to designated (or targeted) positions, the system also demands reliable and accurate landing of the vehicle at nominal and emergency situations. Therefore, ensuring reliable and rapid Received 5 October 2024; revised 8 March 2025; accepted 8 April 2025. Date of publication 23 April 2025; date of current version 2 May 2025. This article was recommended for publication by Associate Editor J. Wang and Editor Z. Li upon evaluation of the reviewers’ comments. This work was supported in part by the National Research Council of Canada’s Artificial Intelligence for Logistics Program, in part by the Natural Sciences and Engineering Research Council of Canada, and in part by the Memorial University of Newfoundland. (Corresponding author: Nirasha Herath.) Nirasha Herath, Oscar De Silva, and George K. I. Mann are with the Intelligent Systems Laboratory, Faculty of Engineering and Applied Science, Memorial University of Newfoundland, St. John’s, NL A1C 5S7, Canada (e-mail: vnherath@mun.ca; oscar.desilva@mun.ca; gmann@mun.ca). Awantha Jayasiri is with the Flight Research Laboratory, National Research Council of Canada, Ottawa, ON K1V 9B4, Canada (e-mail: awantha.jayasiri@nrc-cnrc.gc.ca). Digital Object Identifier 10.1109/TASE.2025.3563822 LZ detection to improve safe autonomous landing and expedite decision-making remains a significant challenge, particularly in unknown and unprepared terrains. Autonomous systems must identify safe LZs with precision and should avoid landing on unsafe areas, such as water bodies, inclined surfaces, or other unstable surfaces. Furthermore, VTOL-based aerial datasets typically generate low-resolution and non-repetitive scan LiDAR point clouds, making accurate LZ detection more challenging. The work proposed in the paper provides a fast, accurate, and semantic-aware LiDAR-based LZ detection that provides a low risk of misidentification, which ensures safe landings in complex environments. The proposed method has been trained and validated using real-world custom-labeled aerial datasets and deployed on reconfigurable hardware, ensuring real-time suitabil- ity for LZ detection in diverse environments. Since the method relies solely on LiDAR sensors, practitioners can implement the proposed method without additional sensor modalities or high- resolution LiDAR point clouds of environments.