This repository contains the source code that use OpenAI API to generate models from implementation code. Give an implementation of UAV application (e.g., C++ code), we generate an intermediate representation (IR) of the code using LLM. Then the IR is again used to generate the model using LLM.

The example directory is a corpus of MAVSDK applications for UAV in C++. Each subdirectory contains the generated IR and corresponding AADL model. We are using gpt-3.5-turbo openai LLM for the generation.

Follow these steps to create a virtual environment.

1. Clone this repository.

2. Run this on command line to create a virtual environment:

python3 -m venv llmenv
source llmenv/bin/activate

Run the following command to update pip on Python: python3 -m pip install --upgrade pip.

3. Navigate to the repository and install required packages:

pip install -r requirements.txt

4. Run the python extract-model examples to generate the AADL models for the MAVSDK code examples. The generated models are stored in the corresponding subdirectories.

Before running the command, place the openai key (api_key.txt) in the root directory of the repository.

How to verify the correctness of generated AADL model:

• Johnsen, Andreas, Paul Pettersson, and Kristina Lundqvist. "An architecture-based verification technique for AADL specifications." European Conference on Software Architecture. Berlin, Heidelberg: Springer Berlin Heidelberg, 2011.

• Johnsen, Andreas, et al. "Automated verification of AADL-specifications using UPPAAL." 2012 IEEE 14th International Symposium on High-Assurance Systems Engineering. IEEE, 2012.

• Ling, Dongyi, et al. "Reliability evaluation based on the AADL architecture model." Journal of Networks 9.10 (2014): 2721.