Fuzzing Robotic Software Using HPC and LLM

This work investigates the application of fuzzing to enhance the cybersecurity of complex robotic software. Fuzzing requires substantial computational resources for automated simulation and testing. Employing generative AI methodologies in the fuzzing process, particularly in message generation and mutation, which enhances the efficacy of these tools. This paper describes a proof of concept by integrates a generalist generative Large Language Model (LLM), marcoroni, into the Robofuzz framework. This integration showcases the potential of AI in fuzzing and provides a ready-to-use deployable tool that can be refined with more targeted models. It is designed for modular utilization with Singularity container technology, facilitating large-scale deployment in High-Performance Computing (HPC) environments. All associated materials, including the proof of concept and code, are accessible on GitHub for review and utilization.