TL;DR
This paper introduces a comprehensive dataset of in-flight positional and energy consumption data from 209 autonomous flights of a DJI Matrice 100 quadcopter performing package delivery simulations. The dataset, collected under varying operational parameters like speed, altitude, and payload, aims to address the critical need for empirical data to improve UAV energy models and advance the understanding of drone delivery operations.
What Problem Does This Paper Address?
The increasing interest in using Uncrewed Aerial Vehicles (UAVs) for last-mile package delivery is hampered by a significant challenge: accurately predicting and optimizing their energy consumption. Multirotor UAVs, unlike fixed-wing aircraft, require substantial energy to generate lift, making range and endurance critical limitations. While theoretical models exist, a recent survey highlighted a wide variation (3-5x) in energy consumption predictions even with common parameters. This discrepancy underscores the urgent need for high-quality, empirical data from comprehensive field tests to validate and refine these models. Existing research often lacks sufficient flight data or fails to release raw data, creating a gap for researchers and industry professionals seeking to improve UAV design, safety, and energy efficiency.
Methods
The researchers developed a robust experimental setup to collect detailed in-flight data. They utilized a DJI Matrice 100 quadcopter equipped with an array of onboard sensors:
- GNSS/INS (3DM-GX5-45): For precise positional, velocity, and orientation data.
- Ultrasonic Anemometer (FT205): To measure wind speed and direction.
- Power Sensor (Mauch PL-200): To record current and voltage for power consumption calculations.
- Raspberry Pi Zero W with ROS: To synchronize and record data from all sensors at approximately 5 Hz.
The data collection involved 209 autonomous flights, each following a predefined triangular route designed to be wind-neutral. Key operational parameters were systematically varied across these flights:
- Commanded Ground Speed: 4, 6, 8, 10, and 12 m/s.
- Cruise Altitude: 25, 50, 75, and 100 m.
- Payload Mass: None, 250 g, and 500 g.
Each combination of these parameters was repeated multiple times to ensure data reliability. The flights were conducted in an open field outside urban areas, adhering to FAA regulations. A detailed pre-flight and post-flight checklist was followed to ensure safety and data integrity.
Datasets
The primary contribution of this paper is the introduction of a new, comprehensive dataset. This dataset comprises:
- Raw ROSBag files: Containing the synchronized sensor data for each of the 209 flights.
- CSV files: Processed data, including flight parameters (duration, wind speed/direction, current, voltage, position, orientation, speed, acceleration) and a file detailing all flight parameters.
This dataset is publicly available, offering a valuable resource for researchers to develop and validate energy consumption models for multirotor UAVs. It differs from existing datasets by its sheer volume (209 flights, ~10.75 hours, ~65 km total distance) and the systematic variation of key operational parameters, providing a rich basis for empirical analysis.
Key Insights and Ideas
- The critical need for empirical data: The paper strongly advocates for the use of real-world flight data to bridge the gap between theoretical energy models and actual UAV performance, highlighting the significant discrepancies in current predictive models.
- Systematic parameter variation: By systematically altering speed, altitude, and payload, the researchers created a dataset that allows for detailed analysis of how each factor influences energy consumption. This controlled approach is crucial for isolating the impact of individual variables.
- Data validation is key: The authors emphasized validating their collected data by comparing flights with similar operational parameters, ensuring consistency and reliability. This rigorous approach builds confidence in the dataset's utility.
- Anemometer bias is a known issue: The paper acknowledges a potential bias in onboard anemometer measurements, particularly at lower UAV speeds, and points to ongoing research to address this. This transparency is important for users of the data.
Key Results
The dataset encompasses 209 flights with a total flight time of 10 hours and 45 minutes, covering approximately 65 kilometers. The data was collected between April and October 2019.
- Energy consumption consistency: Analysis of flights with similar setups showed minor variations in parameters, with results consistently falling within physical boundaries. The mean relative energy amplitude across flight groups was 4.3% with a standard deviation of 2.6%, indicating good repeatability.
- Payload impact: Figure 9 visually demonstrates the impact of payload on power demand, showing a clear increase in power consumption with heavier payloads.
- Altitude and speed variation: Figures 5 and 8 illustrate the programmed altitudes and ground speeds achieved during flights, showing how these parameters were controlled and varied.
- Anemometer correction discussed: While the raw data is provided, the paper includes a section on anemometer bias and ego-motion correction, suggesting methods to improve wind measurement accuracy, though these corrections were not applied to the provided dataset.
Limitations and Future Work
- Anemometer bias: The authors acknowledge that onboard anemometers can exhibit a bias, reporting higher wind magnitudes than expected, especially at lower UAV speeds. While they provide a method for correction, the raw data is released without this correction, leaving it to the user to apply if needed.
- Extrapolation caution: The dataset should not be extrapolated to assess the energy consumption of different drone models or drones operating outside the tested range of parameters.
- Future work on wind correction: The paper suggests that future work is required to investigate and correct the magnitude bias in wind measurements, particularly concerning UAV motion.
- Broader application of data: The authors envision the data being used for various purposes, including modeling energy consumption, validating theoretical models, assessing parameter impacts, and estimating non-measured parameters like drag coefficients.
Summary updated