Synthetic UAV Flight Trajectories

A large set of synthetic UAV flight trajectories produced by the RIOTU Lab's automated ROS 2 / Gazebo / PX4 pipeline (FlightGen). A simulated multirotor tracks time-parameterized reference paths — circles and lemniscates with randomized centres, radii and angular velocities — while its ground-truth 3D position is logged and resampled at 10 Hz. Each trajectory is a standalone CSV of `timestamp, tx, ty, tz`; it was built as training data for the VECTOR GRU trajectory-prediction model. *This is a narrow, low-richness dataset: pure position traces of two analytic path shapes, with no imagery, no other sensors, no actions and no failures. Useful for trajectory prediction, not for robot-learning. The paper claims 5,375 trajectories and 20 h; the actual release contains 5,093 CSVs measuring 21.3 h — the figures here are measured from the released data. The sample images are rendered from the real CSVs, since the dataset contains no imagery of its own.*

A large set of synthetic UAV flight trajectories produced by the RIOTU Lab's automated ROS 2 / Gazebo / PX4 pipeline (FlightGen). A simulated multirotor tracks time-parameterized reference paths — circles and lemniscates with randomized centres, radii and angular velocities — while its ground-truth 3D position is logged and resampled at 10 Hz. Each trajectory is a standalone CSV of `timestamp, tx, ty, tz`; it was built as training data for the VECTOR GRU trajectory-prediction model. *This is a narrow, low-richness dataset: pure position traces of two analytic path shapes, with no imagery, no other sensors, no actions and no failures. Useful for trajectory prediction, not for robot-learning. The paper claims 5,375 trajectories and 20 h; the actual release contains 5,093 CSVs measuring 21.3 h — the figures here are measured from the released data. The sample images are rendered from the real CSVs, since the dataset contains no imagery of its own.*

A large set of synthetic UAV flight trajectories produced by the RIOTU Lab's automated ROS 2 / Gazebo / PX4 pipeline (FlightGen). A simulated multirotor tracks time-parameterized reference paths — circles and lemniscates with randomized centres, radii and angular velocities — while its ground-truth 3D position is logged and resampled at 10 Hz. Each trajectory is a standalone CSV of `timestamp, tx, ty, tz`; it was built as training data for the VECTOR GRU trajectory-prediction model. *This is a narrow, low-richness dataset: pure position traces of two analytic path shapes, with no imagery, no other sensors, no actions and no failures. Useful for trajectory prediction, not for robot-learning. The paper claims 5,375 trajectories and 20 h; the actual release contains 5,093 CSVs measuring 21.3 h — the figures here are measured from the released data. The sample images are rendered from the real CSVs, since the dataset contains no imagery of its own.*

A large set of synthetic UAV flight trajectories produced by the RIOTU Lab's automated ROS 2 / Gazebo / PX4 pipeline (FlightGen). A simulated multirotor tracks time-parameterized reference paths — circles and lemniscates with randomized centres, radii and angular velocities — while its ground-truth 3D position is logged and resampled at 10 Hz. Each trajectory is a standalone CSV of `timestamp, tx, ty, tz`; it was built as training data for the VECTOR GRU trajectory-prediction model. *This is a narrow, low-richness dataset: pure position traces of two analytic path shapes, with no imagery, no other sensors, no actions and no failures. Useful for trajectory prediction, not for robot-learning. The paper claims 5,375 trajectories and 20 h; the actual release contains 5,093 CSVs measuring 21.3 h — the figures here are measured from the released data. The sample images are rendered from the real CSVs, since the dataset contains no imagery of its own.*

simulation · A large set of synthetic UAV flight trajectories produced by the RIOTU Lab's automated ROS 2 / Gazebo / PX4 pipeline (FlightGen). A simulated multirotor tracks time-parameterized reference paths — circles and lemniscates with randomized centres, radii and angular velocities — while its ground-truth 3D position is logged and resampled at 10 Hz. Each trajectory is a standalone CSV of `timestamp, tx, ty, tz`; it was built as training data for the VECTOR GRU trajectory-prediction model. *This is a narrow, low-richness dataset: pure position traces of two analytic path shapes, with no imagery, no other sensors, no actions and no failures. Useful for trajectory prediction, not for robot-learning. The paper claims 5,375 trajectories and 20 h; the actual release contains 5,093 CSVs measuring 21.3 h — the figures here are measured from the released data. The sample images are rendered from the real CSVs, since the dataset contains no imagery of its own.*

Open · Apache-2.0 · custom