A Design Framework for Long-Reach Manipulators in Confined Spaces Using Task-Based Kinematic Optimization and Surface Clustering

Industrial applications involving hazardous materials often require inspection within confined interiors, posing significant challenges due to restricted access and complex geometries. This paper presents a task-driven design synthesis framework for optimizing manipulators for full-surface inspection inside a nuclear fusion vacuum vessel. To facilitate the kinematic optimization pipeline, we introduce a methodology using unsupervised surface clustering of CAD-derived geometries to extract a compact set of reachability targets, combined with a 2D plane projection method for fast collision pre-checks. Using this reduced task set, a multi-objective NSGA-II simultaneously optimizes kinematic type and dimensions to minimize total link length and static joint torques while satisfying reachability and collision constraints. These methods are formed as an integrated pipeline that systematically bridges the gap between abstract and evolving task-space definitions and optimized kinematic synthesis for confined environments. A representative case study of a nuclear fusion vacuum vessel demonstrates that the framework efficiently identifies consistent designs with respect to defined surface inspection metrics. Compared to conventional sampling methods, the proposed approach achieves superior task-space coverage with significantly fewer reachability points. While conventional methods can also yield acceptable results when a sufficiently large number of task points is used, the proposed approach maintains consistent performance even with significantly fewer points, achieving mean runtime reductions of up to 52%. The results demonstrate a systematic and computationally efficient approach to the early-stage design of long-reach manipulators for access-limited confined environments. Although developed for fusion-vessel inspection, the framework can adapt to changes in geometry, constraints, and task requirements.