Generative AI for Reproducible Research in Control, Automation and Robotics

Abstract

Reproducing the results of a research paper is typically difficult and time consuming. Even when the source code is publicly available, the installation and configuration process can fail due to the slight differences in the versions of the software tools and libraries used by the original authors. Software images and containers are a solution to the dependency issues, since they provide the user with a complete environment with all the necessary components for running the software. But the steps for building such containers must be carefully engineered and described in detail. Generative AI tools may provide a valuable help to the human programmer during the development process. In this paper we present a method for building a software image of an open source code repository, which can be readily reproduced by any user. Generative AI is used for processing the textual information in the repository, and producing the building instructions. The AI agent is able to take into account any errors reported by the user in an iterative dialogue until a successful outcome is obtained. The quality of the results is comparable to those produced by a human programmer, demonstrating the validity of the approach.