VisionBasedAircraftDAA

Overview

This repository contains datasets, models, and simulators for the AVOIDDS (Aircraft Vision-based Intruder Detection Dataset and Simulator) benchmark which centers around the vision-based aircraft detect-and-avoid (DAA) problem. The full AVOIDDS dataset of 72,000 samples available here: purl.stanford.edu/hj293cv5980.

Example of Aircraft Detect-and-Avoid in Action!

Primary Features

• Dataset Generation: generate large datasets for training and evaluating object detection model
• Model Training: train custom detection models
• Test Set Evaluation: evaluate models on a test set
• Downstream Task Evaluation: evaluate models on simulated aircraft encounters

Quick Links

• src: Contains the code for the main functionality of the repository.
  • data_generation: Contains the code for generating datasets
  • notebooks: Contains jupyter notebooks for visualizing the outputs of the repository
  • model: Contains code for training and evaluating detection models
  • simulator: Contains code for running encounter simulations between aircraft that use object detection models to issue safety advisories to aircraft
• datasets: Contains subfolders for datasets generated by the data_generation feature of the repository
  • FORMAT.pdf: In-depth description of the format of each dataset outputted via the code in data_generation
• models: Contains baseline object detection models. New models can be generated using the model training functionality

Setting Up X-Plane

1. Download X-Plane 11: Download Link
2. Make sure you are running X-Plane 11.5+. If you are not, follow the prompts on X-Plane to update your version.
3. Download and configure X-Plane Connect: Instructions Here (Make sure you download the file titled XPlaneConnect.zip)
4. In terminal, run pip3 install -r requirements.txt to install the necessary dependencies.