Improve Drone Monitoring of Railroad Tracks by Using a Predictive Flight Control System to Reduce Average Distance Error of the Drone's Footage.
Lessons Learned from Simulation and Field Experiments Evaluating a Vision-Based, Global Positioning System (GPS)-Free Unmanned Aerial Vehicle (UAV) System for Autonomous Railroad Inspection in South Carolina.
Winnsboro
Intelligent Aerial Drones for Traversability Assessment of Railroad Tracks (Year/Phase 2)
Summary Information
Effective inspection and maintenance are essential to ensuring railroad safety and reliability. Traditional railroad inspection methods are often labor-intensive, time-consuming, and costly. To improve inspection efficiency while reducing operational costs, this project explores the use of UAVs to assist railroad inspection, especially for post-event inspection and traversability assessment following extreme events such as floods, hurricanes, and earthquakes. Building on the track detection algorithm developed in Phase 1, Phase 2 of this project advanced both the perception and flight control components to enable a fully autonomous, vision-based UAV track-following system using computer vision. This method eliminated reliance on external positioning sensors such as GPS, making it well suited for GPS-degraded or denied locations. Simulation experiments were conducted at a university campus and outdoor flight tests were performed along multiple railroad segments at the South Carolina Railroad Museum in Winnsboro, South Carolina.
The experiments were conducted under multi-track conditions (e.g., parallel track sections and curved alignments) to evaluate the proposed system’s track detection and flight control performance. The track following performance was assessed by comparing the UAV’s flight trajectory with ground-truth railroad track coordinates. The experimental results provide several lessons learned related to both algorithm development and system design.
- Improve drone monitoring of railroad tracks by using a predictive flight control system to reduce average distance error of the drone's footage. Replacing reactive proportional–integral–derivative (PID) control with more advanced control algorithms such as model predictive control (MPC) to provide a strong foundation for collision-aware navigation and reduce average distance error in autonomous inspection missions. These approaches leverage system dynamics and short-horizon future-state Replacing reactive proportional–integral–derivative (PID) control with more advanced control algorithms such as model predictive control (MPC) to provide a strong foundation for collision-aware navigation and reduce average distance error in autonomous inspection missions. These approaches leverage system dynamics and short-horizon future-state prediction to generate optimal motion commands while explicitly accounting for multivariable constraints to achieve smoother path following with improved responsiveness.
- For GPS-degraded or denied locations, design GPS-free UAV inspection systems through careful device configuration and system integration. Key components include an onboard mini-PC to support real-time algorithm execution; vision sensors with built-in simultaneous localization and mapping (SLAM) to enable GPS-independent positioning; optical flow and proximity sensors to support stable flight and obstacle avoidance; and tightly integrated hardware–software architectures (e.g., Robot Operating System (ROS)-based control frameworks) to achieve autonomous operation without excessive computational overhead.
