Terrain following - low level fixed altitude flying

Abstract

This thesis studies a terrain following capability for fixed-wing tier 1 UAS (<25 kg), designing a system to perform very low-level altitude flights adapting to the terrain contour and avoiding obstacles. It is the first work on terrain following combining uncertain digital elevation models (DEM) and real-time observations with risk-awareness. An innovative dual-stage system is proposed, using a cubic B-spline curve to generate an upper envelope combining DEM datasets and direct ground mapping measurements through onboard sensors, and a non-linear model predictive control (NMPC) to track the reference envelope with altitude range constraints. The system is designed for real-time implementation, employing moving window predictions, and an aggressiveness modulation to improve solver times while safely overcoming obstacles. The cubic B-spline DEM-Obstacle envelope is a geometric object that is found through solving a quadratic program with guaranteed convergence. The NMPC uses the full non-linear longitudinal dynamic model of the UAS to provide optimal vertical guidance and control to the non-linear underactuated platform, tracking the envelope. The performance is critically sensitive to the rangefinder angular uncertainty, forcing higher flight paths while maintaining minimal collision risk. Chance constraint formulation in the envelope allows improvements through moderate risk allowance, balancing a trade-off between risk and performance. Although the obstacle avoidance sensors are essential, the best performance is achieved using both a quality elevation dataset and sensor suite, employing LiDAR DEM and LiDAR rangefinders.