Drone onboard multi-modal sensor dataset for complex outdoor scenarios

The Data acquisition missions were designed and executed using DJI Pilot 2’s flight route planning feature. The missions encompassed five distinct geometric patterns: 1. triangular, 2. circular, 3. rectangular, 4. linear, and 5. multi-dimensional. Each mission was configured as a waypoint flight path, allowing precise customization of parameters such as altitude, speed, and turning angle for each waypoint. The dataset consists of 3D space flight data such as take-off, landing and varying altitude to introduce the z-axis changes. It must be noted that data was logged at a frequency of 10 Hz.

To ensure consistency within the data, identical parameters were maintained across all data acquisition missions. The dataset comprises 20 distinct flights, with each flight path repeated multiple times, resulting in approximately 30 minutes of flight time per mission. The dataset is structured as time-series data, with each flight uniquely identified by a flight number and corresponding timestamp. The drone's spatial position is represented by the variables position_x, position_y, position_z  while its orientation is captured by the variables orientation_x, orientation_y, orientation_z, orientation_w.  Additionally, the drone's velocity and angular velocity are represented by the variables velocity_x, velocity_y, velocity_z, angular_x, angular_y, angular_z respectively. The linear acceleration is described by the variables linear_acceleration_x, linear_acceleration_y, linear_acceleration_z. The dataset also includes environmental data such as wind_speed, wind_angle using the TriSonica Mini Wind and Weather Sensor as well as information regarding the drone's battery status, including battery_voltage, battery_current.

Data Acquisition Paths: Data acquisition paths

The dataset includes labels for various operational states of the drone, such as IDLE_HOVER, ASCEND, TURN, HMSL and DESCEND. These labels can be utilized to classify the drone's current activity. Moreover, the annotated dataset can be applied in multi-task learning to predict the drone's trajectory.

The DJI Matrice 300 RTK is utilized as the primary platform for data acquisition, leveraging its compatibility with onboard development kits to facilitate the extraction of data from its integrated sensors and flight controller. To execute the developed software the NVIDIA Jetson Xavier NX serves as the embedded computing device. Utilizing the  Onboard software development kit the Jetson Xavier NX enables real-time access and processing of data from the drone's sensors and flight controller.