Remote Maintenance Path Planning Methods for Plant Architecture Assessments

Remote Maintenance (RM) path planning of In-Vessel Components (IVC) informs the early design decisions on the Remote Maintenance system and on the DEMO plant architecture. The complexity of the motion planning problem is emphasized due to the holistic approach needed in the DEMO project as path planning is part of the RM design which has interdependencies with all of the interfacing systems. Path planning in this context has historically been executed with manual step by step path generation and clearance measurement with CAD tools. A review of existing tools available for path planning is executed to explore the various ways of pursuing a more automated approach in generation of in-vessel RM paths for IVCs. The goal of the work is an increase in speed and a more harmonized quality in the tasks related to paths, and to pave way for a more converged approach in the future.
To account for the specific criteria related to path feasibility a special use case is constructed as an experiment. Catia is used as an example of a CAD tool utilization for path planning, and as a final step game-engine Unity is utilized to study the possibilities of Machine Learning Methods in path planning for DEMO. From Machine Learning paradigms, Reinforcement Learning (RL) is used for its capability to perform in an interactive simulated environment. The environment is a Vacuum-Vessel sector, and the interactive agent is a Blanket segment. The approach provides dynamic adaptability to changes in the tokamak environment to assess different plant design points in the future.