Augmenta Tractor-Drone Co-Robotics Dataset for Weed Detection

📌 Overview This dataset contains telemetry and computer vision metrics collected from a Smart Agriculture co-robotics system developed by Augmenta (acquired by CNH Industrial). The system consists of a tractor equipped with a "Field Analyzer" and an autonomous drone (UAV). The data was collected to train Machine Learning models (specifically XGBoost) to predict the should_fly event—a signal that triggers the drone to launch and assist the tractor when the tractor's onboard cameras are blinded by environmental factors (e.g., sun glare/lens flare). This work was conducted as part of the MLSysOps project (EU Horizon Europe). 🚜 System Context The Augmenta system automates the application of fertilizers and herbicides using Real-Time Computer Vision. 1. Normal Operation: The tractor cameras detect weeds and spray precisely. 2. The Problem: When the sun is at a specific angle (e.g., sunset/sunrise), it creates lens flare, blinding the tractor's camera ("Sensor Fault"). The system enters "Safe Mode" and sprays the whole field blindly, wasting chemicals. 3. The Solution: The system predicts this fault and deploys a Drone to fly ahead of the tractor. The drone sends clear weed detection coordinates back to the tractor, allowing precise spraying to continue. 📂 Data Dictionary The dataset consists of time-series telemetry. The core goal is to predict should_fly using the sensor and performance metrics. Column Name Type Description timestamp datetime ISO 8601 Timestamp of the recording. quality_indicator_1 int Confidence metric: Number of data correspondences between samples. quality_indicator_2 int Confidence metric: Number of data points used for localization. field_indicator_1 int The number of detected weeds in the current frame. field_indicator_2 float Fraction of the field frame under environmental variation. sensor_fault_probability_1 float Key Feature: Probability of the camera sensor being blinded by the sun (0.0 to 1.0). environment_sensor_1 float Ambient light/environmental condition measurement. processing_performance float Average processing speed/performance metric of the vision unit. success_rate float Fraction of successfully processed image frames (0.0 to 1.0). should_fly int Target Variable: Binary flag (0 or 1). 1 indicates the drone should be deployed. heading float Instant heading of the vehicle (radians). velocity float Instant velocity of the vehicle (m/s). latitude float GNSS Latitude. longitude float GNSS Longitude. altitude float GNSS Altitude (meters). time_since_sensor_fault float Time (seconds) elapsed since the last sensor fault 📊 Statistical Summary Descriptor Count Mean Std Min 25% 50% (Median) 75% Max quality_indicator_1 1053 545.06 145.27 9.0 454.0 531.0 631.0 1051.0 quality_indicator_2 1053 413.88 119.38 64.0 324.0 396.0 492.0 835.0 field_indicator_1 1053 52.12 58.25 0.0 18.0 37.0 66.0 767.0 field_indicator_2 1053 0.015 0.010 0.001 0.008 0.013 0.021 0.086 sensor_fault_prob_1 1053 0.185 0.271 0.000 0.0002 0.086 0.384 0.999 environment_sensor_1 1053 10187.9 8416.8 808.5 2456.2 8959.3 20497.2 25678.0 processing_perf 1053 12.48 1.98 4.99 10.48 13.03 14.44 15.52 success_rate 1053 0.51 0.47 0.00 0.00 0.57 1.00 1.00 should_fly 1053 0.37 0.48 0.0 0.0 0.0 1.0 1.0 heading 1053 0.81 1.66 -3.13 -0.66 1.34 2.48 3.13 velocity 1053 2.72 0.99 0.01 2.22 2.78 3.33 5.25 altitude 1053 276.11 38.97 254.2 255.7 270.9 273.5 443.5 🧪 Collection Methodology Location: Perivlepto, Volos, Greece (Augmenta Test Field). Conditions: Data was specifically collected during sunset/sunrise to induce lens flare and trigger the "Safe Mode" (sensor fault) scenarios. Equipment: Tractor Node: Standard agricultural tractor with Augmenta Field Analyzer (Cameras + Edge Compute). Drone Node: Custom UAV integrated with the Augmenta control stack. Protocol: The tractor performed "Back-and-Forth" scanning of the field. As the tractor turned into the sun, the sensor_fault_probability spiked, triggering the should_fly signal for the drone. 📜 Citation If you use this dataset in your research, please use the citation generated by Zenodo (located in the right sidebar of this record). 🤝 Acknowledgements & Funding This work is part of the MLSysOps project, funded by the European Union’s Horizon Europe research and innovation programme under grant agreement No. 101092912. More information about the project is available at https://mlsysops.eu/