Autonomous Copilot Agent for Flight Simulation

This project simulates an autonomous copilot agent that collaborates with a human pilot in a flight simulator (X-Plane 11) during the take-off phase, particularly managing abnormal situations like engine failure after a bird strike. The agent uses a combination of machine learning (Random Forest classifier) and cognitive modeling (ACT-R) to determine the level of support needed by the pilot.

Folder Structure

• data/: Contains the logged data from the flight simulations.

  • 'cool' in the filename refers to expert pilot data.
  • 'nul' in the filename refers to novice pilot data.

• xpc/: The NASA XPlaneConnect plugin for setting up UDP communication between Python and the X-Plane 11 flight simulator.

• actr.py: A Python wrapper for interacting with the ACT-R cognitive architecture.

• annunciator.py: A utility script for aural annunciations. This script listens to a queue from the main script (engine_failure.py) and returns the updated queue with messages, which are then executed on the main thread.

• classifier.py: Contains the logic for real-time classification using the Random Forest model. It is called by the main script to send flight log data and returns a classification (0 = novice, 1 = expert).

• engine_failure.py: The main script that sets up the ACT-R environment, initiates ACT-R experiments, and manages threads for other scripts.

• engine_failure_model_mcgill_ai.lisp: The ACT-R cognitive model written in Lisp, defining the memory and production rules for the agent.

• log_takeoff.py: A script for logging flight data from X-Plane 11 every 10 knots.

• flightdir.py: A script for displaying the flight director at 10° pitch angle during the takeoff.

• manual_data_comparison.py: A script for manually comparing two .csv files (e.g., "expert.csv" and "novice.csv") stored in the root folder.

• random_forest_model_joblib: The Random Forest model output from Scikit-learn (joblib format).

• random_forest_model.pkl: The Random Forest model output from the custom implementation (pkl format).

• random_forest_scikit_training.py: Random Forest Classification for T.A.R.S. Train a Random Forest classifier using Scikit-Learn Machine Learning library to predict flight data based on various features, used for comparison with our own implementation.

• random_forest_training.py: Our own Random Forest Classification for T.A.R.S. Train a Random Forest classifier to predict flight data based on various features. The classifier is implemented using decision trees, and it performs cross-validation to evaluate its performance. The script allows to experiment with different hyperparameters, including the number of trees, maximum depth, the minimum samples required to split a node, and cross-validation folds. Additionally, it provides the ability to experiment with different random seeds for reproducibility.

Requirements

• Python 3.x
• Required Python libraries:
  • sklearn (for machine learning)
  • numpy, pandas, matplotlib (for data manipulation and visualization)
  • xpc (for X-Plane integration)
  • joblib (for loading the Random Forest model)
  • pickle (for loading the Random Forest model)
  • pyttsx3 (for speech synthesis of the announces)

Acknowledgements

This software uses scikit-learn, which is licensed under the BSD 3-Clause License.