Sensor Model-Based Trajectory Optimization for UAVs Using Nonlinear Model Predictive Control
Markus Zwick, Matthias Gerdts, Peter Stütz · AIAA SCITECH 2022 Forum · 2022
View Video Presentation: https://doi.org/10.2514/6.2022-1286.vid In this paper, we present an approach to enhance the overall detection performance in the field of unmanned aerial reconnaissance with imaging sensors. We achieve this task by generating optimized reference trajectories for UAVs through a recurring process of sequential path planning and trajectory optimization with nonlinear model predictive control. These trajectories are setpoints for the movement of the UAV in a spatio-temporally varying potential field, which represents the detection performance. We utilize known sensor performance models to map the sensor position dependent detection performance under the impact of numerous influencing factors. These are perceptual, sensor platform and environmental aspects as well as mission objectives. Our approach can be briefly summarized as follows: first, a discrete sensor footprint planning is performed via coverage path planning in the area to be reconnoitered. For each footprint, a perception map is created, which is a state-dependent representation of the detection performance in the UAV's flight plane. The environmental state vector is mapped to the perception map via the sensor performance model. Next, in a cyclic manner, a fan-shaped path planning is conducted starting from the current UAV position. Along these paths, the spatio-temporal detection performance is determined utilizing the perception maps. The cost for trajectory generation is then calculated for each path element using nonlinear model predictive control. A downstream evaluation of the path-dependent detection performance and the associated trajectory cost is used to determine the optimal path, which serves as a setpoint when generating the next section of the reference trajectory. This process is repeated until the end of the recon area is reached. We evaluate our approach conducting a series of simulated reconnaissance tasks for vehicle detection with a fixed wing UAV. In an area reconnaissance scenario, our approach shows an improvement in detection performance of about 5% compared to a primitive trajectory benchmark. Even for a route reconnaissance scenario, a slight improvement of about 0.6% compared to an already very high absolute benchmark value is measured.