UAV multimodal trajectory prediction framework for ProjectAirSim
The project implements a full pipeline to predict multiple future UAV paths from past kinematics, using a GRU‑based model inspired by TNT/DenseTNT. It includes data adapters, training, inference, visualization, and an online node that plugs into the ProjectAirSim simulator. Designed for researchers and developers building behavior models, adversarial testing, or closed‑loop control validation for drones, it offers end‑to‑end code rather than just a concept. Its tight integration with ProjectAirSim and support for both offline and live prediction set it apart from generic trajectory libraries.
View on GitHub →QinCheng0928/Projectairsim-UAV-Multimodal-Trajectory-Prediction