Paper Review: PL-ALF Enables UAV Autonomous Flight in Low-Texture Environments

PL-ALF system framework

The PL-ALF framework is mainly composed of point-line feature fusion SLAM module and path optimization module. The SLAM system uses binocular cameras and IMU sensors, combined with loop detection and local/global BA optimization, to provide high-precision position information. The path optimization module uses the A-Star algorithm for global path planning and combines it with L-BFGS for trajectory smoothing.

Point and line feature fusion SLAM

• The ORB + LSD (line segment detection) method is used to improve feature extraction capabilities in low-texture environments.

• Combined with local/global bundle adjustment (BA) for optimization to enhance localization accuracy.

• Fusion of IMU data improves system robustness and reduces errors caused by feature loss.

path planning and obstacle avoidance optimization

• Combined with visual SLAM localization and depth camera information, UAV trajectories are updated in real time.

• A-Star algorithm + L-BFGS optimization, only corrects paths with obstacles to improve calculation efficiency.

• B-Spline smooth trajectory generation is used to ensure the stability and feasibility of the flight path.

Simulation + real machine verification

Experimental platform

▪ hardware:AMOVLABP230 quadcopter UAV, equipped with Intel Realsense D435i depth camera, KV 1750 motor, Jetson Xavier NX processor, and Pixhawk 4 Miniflight controller.

▪ software: Prometheus open-source platform

Simulation experiment

▪ Testing the SLAM localization accuracy on the EuRoC data set, PL-ALF improves the localization accuracy by 30\% compared to ORB-SLAM3.

▪ In the simulation environment, the path planning capabilities of PL-ALF were tested and compared with multiple mainstream algorithms (Fast-Planner, EGO-Planner, EWOK). Experimental results show that PL-ALF has a higher success rate of obstacle avoidance and is more stable than other methods, especially in low-texture environments.

Real machine test

To further verify the autonomous obstacle avoidance capability of PL-ALF, the research team used AMOVLABP230UAV to conduct tests in low-texture, narrow indoor corridors, and successfully completed autonomous flight tasks (such as passing through door frames, avoiding walls, etc.). The narrowest passage width was less than 1 meter, which fully demonstrated the adaptability of PL-ALF in complex environments.

Low texture corridor localization

Real corridor obstacle avoidance

Paper details

theme:PL-ALF: A Novel Point-Line Feature Autonomous Localization and Flight Framework Based on Multi-sensor Fusion and Optimization PL-ALF: A Novel Point-Line Feature Autonomous Localization and Flight Framework Based on Multi-sensor Fusion and Optimization
Journal:IEEE Transactions on Instrumentation and Measurement
DOI: 10.1109/TIM.2024.3522670
Paper link:https://ieeexplore.ieee.org/document/10816121