| | --- |
| | license: mit |
| | language: |
| | - en |
| | base_model: |
| | - google/flan-t5-small |
| | --- |
| | # Open Rotor Copilot: Neural Blackbox Analysis |
| |
|
| | ## Model Card for `rbarac/open-rotor-copilot` |
| |
|
| | --- |
| |
|
| | ## Model Details |
| |
|
| | **Model Description** |
| |
|
| | Open Rotor Copilot is an open-source neural model for automated blackbox analysis and issue detection in FPV drone flight logs. Given a windowed feature string (representing a segment of Blackbox telemetry), it predicts likely flight anomalies or faults and provides concise human-readable explanations. This model enables pilots, engineers, and researchers to automate log review and root cause analysis for multirotor platforms. |
| |
|
| | - **Developed by:** Bahadir Arac |
| | - **Model type:** Transformer (sequence classification, text-to-label) |
| | - **Language(s) (NLP):** English |
| |
|
| | --- |
| |
|
| | ## Model Sources |
| |
|
| | - **Repository:** [https://huggingface.co/rbarac/open-rotor-copilot](https://huggingface.co/rbarac/open-rotor-copilot) |
| | --- |
| |
|
| | ## Uses |
| |
|
| | ### Direct Use |
| |
|
| | - Automated analysis of Betaflight/INAV/ArduPilot blackbox logs |
| | - FPV drone diagnostics for issues such as vibration, motor problems, signal loss, or sensor faults |
| | - Assisting pilots in understanding potential causes for in-flight anomalies |
| | - Integrating into post-flight dashboards or tools (e.g., Gradio demo, local scripts) |
| |
|
| | ### Downstream Use |
| |
|
| | - Research into drone reliability, flight risk, and root cause modeling |
| | - Educational purposes (for teaching blackbox log interpretation) |
| | - Automated labeling for LLM training |
| |
|
| | ### Out-of-Scope Use |
| |
|
| | - Direct control of flight hardware (this model is for analysis only) |
| | - Use in domains outside of drone telemetry without further evaluation |
| |
|
| | --- |
| |
|
| | ## Bias, Risks, and Limitations |
| |
|
| | - Model accuracy depends on the representativeness of training data |
| | - Unusual or novel failure modes not present in data may not be detected |
| | - Not a replacement for domain expertise—should be used as an assistant, not sole authority |
| |
|
| | --- |
| |
|
| | ## Recommendations |
| |
|
| | - Always review results, especially for safety-critical applications |
| | - Contribute new log data or open issues to help improve the model |
| |
|
| | --- |
| |
|
| | ## How to Get Started with the Model |
| |
|
| | ```python |
| | from transformers import pipeline |
| | |
| | pipe = pipeline("text-classification", model="rbarac/open-rotor-copilot") |
| | |
| | # Example input (windowed feature string, hybrid format) |
| | input_str = "gyro_std=123.56, rcCommand3_mean=1150.2, vbat_drop=0.91, rssi_min=17.2, ..." |
| | |
| | result = pipe(input_str) |
| | print("Predicted Issue(s):", result[0]['label']) |
