TII-RATM Drone Racing#

Indoor drone racing flights pairing onboard visual-inertial odometry with motion capture ground truth, as native .mcap episodes.

A quadrotor flies laps of a four-gate track, three on an ellipse and three on a lemniscate, carrying a fisheye camera and a 500 Hz IMU while a motion capture system watches the room.

Each episode carries both numbers: the estimate the drone computed from its own camera and IMU, and the reference the capture system measured at the same instant. The distance between them is published as a per-pose series, so the error is measurable rather than assumed.

Six flights, 10.9 minutes and 2,563 metres flown, over 17,305 camera frames, 327,086 IMU samples, 293,219 odometry poses and 179,283 capture poses. Tracking error ranges from 0.61 m to 1.66 m RMSE across the six.

Details

Example usage

 1import fiftyone as fo
 2import fiftyone.zoo as foz
 3
 4dataset = foz.load_zoo_dataset("tii-ratm-drone-racing")
 5
 6# The flights the odometry found hardest
 7view = dataset.sort_by("tracking_error_rmse_m", reverse=True)
 8
 9# The lemniscate laps
10view = dataset.match({"track": "lemniscate"})
11
12session = fo.launch_app(dataset)
tii-ratm-drone-racing