In-Vessel Component Remote Maintenance path planning methods for plant architecture assessments

Remote Maintenance (RM) path planning for In-Vessel Components (IVC) provides input for the early design decisions on the RM system and the architecture of fusion power plants. In fusion engineering, path planning is complex, as it is linked with various interfacing systems within the RM design. Traditionally, path planning in this field is performed manually, involving step-by-step path generation and clearance measurements using CAD tools. This study aims to explore existing tools for path planning and develop a novel, a more automated approach to generating in-vessel RM paths for IVCs. The primary goals are to increase the speed of path assessments, improve task consistency, and lay the foundation for a more unified path planning process in the future. To assess path feasibility, a specific use case has been constructed as an experiment. Catia, a CAD tool, is used for manual path planning, while the game engine Unity is utilized to investigate Machine Learning (ML) methods for path planning in fusion devices. Reinforcement Learning (RL), known for its capabilities in interactive environments, is employed in Unity via ML-Agents plugin. The environment is a Vacuum-Vessel (VV) sector, where the interactive agent is a Blanket segment. The RL approach is expected to adapt dynamically to changes within the tokamak environment, allowing for assessments of various plant design points in the future. • Assessment of tools for planning of IVC RM paths • Demonstration of definition and analysis methods for IVC RM paths with CAD tools • Process for ML-Agents plugin in Unity for reinforcement learning in IVC context • Machine learning as a transformative solution for automating path planning in fusion