Reliable distal force feedback is essential for safe guidewire manipulation in robot-assisted endovascular interventions. However, fiber Bragg grating (FBG) sensors suffer from inherent temperature–strain cross-sensitivity, and blood-flow-induced thermal disturbances further introduce measurement drift under realistic intraoperative conditions. These effects lead to inaccurate temperature compensation and biased distal force estimation, thereby degrading the stability and reliability of force feedback.
This study proposes a causal machine learning–based temperature compensation framework for dual-FBG sensing. A thermo–fluid–solid coupled guidewire model is established to describe the interaction among blood flow, structural deformation, and temperature variation, providing a physics-informed representation of thermal–mechanical coupling. Structural causal modeling is introduced to characterize the causal relationships between temperature variation, sensor signals, and mechanical strain while accounting for latent confounding factors. Counterfactual inference is employed to estimate strain under temperature intervention, enabling effective separation of thermal disturbances from mechanical effects. Heterogeneous treatment effects under different operating conditions are captured using a causal forest model, and an online adaptive learning strategy based on Follow-The-Regularized-Leader (FTRL) is adopted to improve robustness under dynamic environments.
In vitro experiments were conducted on a vascular intervention simulation platform at flow velocities of 5, 10, 15 and 20 cm/s. Compared with the conventional dual-FBG method, the proposed approach reduces temperature-compensation mean absolute error by more than 50\% and decreases distal force estimation error by up to 70\%, while suppressing performance degradation at higher flow velocities.
The proposed framework improves the stability and consistency of distal force feedback in dynamic vascular environments, enabling more reliable assessment of guidewire–vessel wall interaction and enhancing the safety of robot-assisted endovascular interventions.