Pipeline Artifacts and Model-Driven Transformations for Robot Script Generation

Overview This repository contains the model transformation scripts and pipeline artifacts associated with the study “MetaRPA: A Model-Driven Framework for Robotic Process Automation.” The repository includes the ATL and Acceleo transformations used in the Model-Driven Engineering (MDE) pipeline, as well as the artifacts generated across its different stages, illustrating how models are progressively transformed into executable robot scripts. The process models representing the case studies used as input for the pipeline are available in the companion dataset containing the BPMN models derived from enriched UI logs (https://doi.org/10.5281/zenodo.19000103). These materials are provided to support transparency and reproducibility of the proposed approach by exposing both the transformation logic and the artifacts produced across the pipeline. Repository Structure The repository is organized into two main directories. The directory 01_M2M_M2C_scripts contains the model transformation specifications used in the pipeline, while 02_Generated_models_&_robot_scripts contains the artifacts generated during the execution of the pipeline for each case study and target platform. Transformation Scripts The directory 01_M2M_M2C_scripts contains the transformation specifications implementing the two stages of the MDE pipeline. The Model-to-Model (M2M) transformations are implemented using ATL and convert the Selected Candidate Routine (SCR) into an Executable Routine (ER). In this stage, the process model is interpreted to derive the execution logic of the robot. The transformations analyze the logical structure of the process, resolve the ordering of tasks and conditional branches, and generate the corresponding robot execution structures such as sequential blocks, conditional branches, and UI interactions. The M2M transformation logic is implemented through the following ATL files: bpmn2robot.lib.atl – auxiliary transformation library providing helper functions used for process structure interpretation, normalization, and execution ordering. bpmn2openrpa.atl – transformation generating ER models targeting the OpenRPA robot metamodel. bpmn2uipath.atl – transformation generating ER models targeting the UiPath robot metamodel. The Model-to-Code (M2C) transformations are implemented using Acceleo templates. In this stage, the ER models generated during the M2M phase are translated into executable robot scripts compatible with the target RPA platforms. The templates interpret the execution structures defined in the ER and materialize them as platform-specific workflow elements required for execution. The repository includes the following templates: openrpa2xaml.mtl – generates robot scripts compatible with the OpenRPA platform. uipath2xaml.mtl – generates robot scripts compatible with the UiPath platform. Both templates take ER models as input and produce .xaml workflow files that can be imported into the corresponding RPA tools. Generated Pipeline Artifacts The directory 02_Generated_models_&_robot_scripts contains the artifacts produced during the execution of the pipeline for the case studies used in the experimental evaluation. Artifacts are organized first by target RPA platform (OpenRPA and UiPath) and then by case study, corresponding to the three evaluation scenarios analyzed in the study: C1 – Invoice Management C2 – Application Evaluation C3 – Unsubscription Process For each case study and platform, the repository includes the artifacts generated across the different stages of the pipeline. These artifacts correspond to the intermediate and final representations used during the transformation process. The following artifact types are included: SCR (.xmi) – Selected Candidate Routine. This model represents the parsed BPMN process with UI interactions and serves as the baseline process representation used as input for the transformation pipeline. Order (.xmi) – Execution Order model. This auxiliary model encodes the execution ordering and branching relationships derived from the process structure and is used to guide the transformation rules during the M2M phase. ER (.xmi) – Executable Routine. This model is produced by the M2M transformation and captures the executable structure of the robot, including control flow, UI interactions, and decision logic. RS (.xaml) – Robot Script. This artifact is generated during the M2C phase and corresponds to the final executable automation workflow that can be imported into the corresponding RPA platform. Naming Convention Generated artifacts follow the naming pattern [prefix]_[case]_[artifact], where each element identifies the platform, case study, and pipeline artifact. The prefix indicates the target RPA platform: op for OpenRPA and uip for UiPath. The case identifier corresponds to the evaluated case study: c1 (Invoice Management), c2 (Application Evaluation), and c3 (Unsubscription Process). The artifact identifier specifies the pipeline artifact: scr (Selected Candidate Routine), order (Execution Order), er (Executable Routine), and rs (Robot Script). For example, the file op_c1_er.xmi represents the Executable Routine generated for case study C1 targeting the OpenRPA platform.