Data-Driven Target Tracking Methods of UAS/UAM in Dynamic Environment
Rachit Prasad, Gwonyeol Lee, Jae-Young Choi, Junki Shim, Nicholas C. Song, Seongim Choi · AIAA SCITECH 2023 Forum · 2023
View Video Presentation: https://doi.org/10.2514/6.2023-2660.vid The unmanned aerial system (UAS)/urban air mobility (UAM) aircraft market is expected to expand rapidly in the coming years. As this happens, these UASs will be expected to operate in increasingly dense and dynamic environments. It would be increasingly important to be able to know the current state of the environment, predict the future state and take decisions accordingly. This included knowing the current position and future trajectory of other UASs in nearby vicinity. This is done conventionally using Kalmon filter-based target tracking approaches, such as extended Kalman Filter (EKF) and Interacting Multiple Model (IMM) based models. While these prediction models work reasonably, they have inherent shortcomings such as being constraint to a discrete number of motion models and inability to consider environment related information such as nearby obstacles. In this study, multiple data-driven target tracking models have been proposed and their performance has been verified by comparing it with EKF and IMM based models. Based on requirements, (1) Gaussian Process Regression (GPR), (2) Long Short-Term Memory (LSTM) encoder decoder neural network, (3) Occupancy map-based LSTM-CNN neural network, and (4) conditional generative adversarial network (CGAN) based models have been proposed and developed. For the training of the data-driven models, trajectory datasets were prepared from realistic UAV scenarios with ~60000 datapoints. A similar validation dataset was prepared, and the data-driven and traditional models were validated on it. The results showed that data-driven methods outperformed the traditional target tracking methods under multiple scenarios.