Koopman-Based Dynamic Environment Prediction for Safe UAV Navigation

TL;DR

The paper tackles safe UAV navigation in dynamic environments with moving obstacles using real-time LiDAR data. It introduces a Koopman operator-based predictor to linearize obstacle dynamics from observation history and integrates this into a convex MPC with linear obstacle constraints. Key contributions include the first integration of Koopman-based dynamic environment modeling in autonomous UAV navigation, demonstration in ROS2-Gazebo, and benchmarking against Gaussian Process-based predictions, showing improved accuracy and substantial computational savings. The results indicate practical viability for real-time deployment in urban or disaster scenarios, enabling safer, more reliable autonomous flight.

Abstract

This paper presents a Koopman-based model predictive control (MPC) framework for safe UAV navigation in dynamic environments using real-time LiDAR data. By leveraging the Koopman operator to linearly approximate the dynamics of surrounding objets, we enable efficient and accurate prediction of the position of moving obstacles. Embedding this into an MPC formulation ensures robust, collision-free trajectory planning suitable for real-time execution. The method is validated through simulation and ROS2-Gazebo implementation, demonstrating reliable performance under sensor noise, actuation delays, and environmental uncertainty.

Figures (8)

• Figure 1: Polygonal obstacle approximation and minimum distance constraint visualization.
• Figure 2: Simulated quadcopter.
• Figure 3: Simplified interaction diagram of the ROS2 nodes relative to the main vehicle control.
• Figure 4: Prediction performance on circular and infinity-shaped trajectories.
• Figure 5: Effect of prediction step and history length on prediction accuracy.