Physics-informed bilevel coordinated planning for robot-assisted transcranial magnetic stimulation
Planning for robot-assisted transcranial magnetic stimulation (TMS) in confined treatment spaces involves the coupled optimization of the base position, redundant axial rotation of the coil, and manipulator trajectory. Directly solving this joint optimization problem is challenging because of its high-dimensional search space and the frequent evaluation of physical constraints. To address these challenges, a bilevel collaborative planning method based on physics-informed bi-stage multi-objective differential evolution (PIBF-MODE) is proposed. According to the distinct roles and timescales of configuration decisions and motion parameters, the planning problem is decomposed into base–terminal configuration optimization and safe path time parameterization. At the upper level, end-effector stiffness and safety-clearance information guide the configuration search, whereas at the lower level, the trajectory time law along the geometric path is determined subject to a dynamic safety envelope. To reduce the computational cost of stiffness evaluation during optimization, a physics-informed stiffness surrogate model is developed, achieving a mean absolute percentage error of 1.8614% for the three-axis translational stiffness magnitude. Experiments on 30 targets covering four typical treatment regions show that the proposed method completes the planning tasks under both fixed-base and free-base conditions. With a fixed geometric path, the lower-level path-velocity parameterization required an average computation time of 0.10 s, whereas the complete bilevel planning under free-base conditions required an average computation time of 229.17 s. In addition, the proposed method maintained planning feasibility under the prescribed perturbations to the planning inputs. The results demonstrate that the proposed method reduces the coupling of the planning problem by hierarchically separating configuration search from path time parameterization, while coordinating robot motion performance and end-effector stiffness and consistently generating multi-target planning solutions that satisfy safety constraints.
Authors
- Ting Sun (ORCID: https://orcid.org/0000-0002-8747-8909)
- Shijun Li
- Peng Niu
- Qiang Cheng
- Zhifeng Liu
- Xiaolong Hao
- Jingjing Xu
- Jie Liu
Institutions
- Hebei University of Technology (CN)
- Chinese PLA General Hospital (CN)
- Beijing University of Technology (CN)
- Beijing Machine Tool Research Institute (CN)
Publication Details
- Journal
- Journal of Computational Design and Engineering
- Published
- 2026-11-20
- DOI
- https://doi.org/10.1093/jcde/qwag102
- Primary Topic
- Robotic Mechanisms and Dynamics
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China
- Beijing University of Technology
- Jilin University
- Beijing Nova Program