Robotic Optical Coherence Tomography-Guided Geometric Modeling of Soft Tissue: 3D Representations for Data-Driven Closed-Loop Shape Control

TL;DR

This paper presents an OCT-guided, data-driven framework for closed-loop soft-tissue shape control using 3D volumetric representations learned from demonstrations. It evaluates multiple representations—full-volume, sparse-volume, and voxelized point clouds—with residual conditioning and both deterministic and diffusion policies, showing improved accuracy and robustness across tasks and tissue types.

Why It Matters

Accurate, real-time control of soft tissues is critical for minimally invasive robotic surgery, especially at microscale where visibility and precision are limited. This work enables reliable, adaptive shape control using high-resolution OCT feedback, paving the way for autonomous or semi-autonomous surgical interventions.

Key Findings

• Volumetric OCT representations enable effective data-driven shape control in both deterministic and generative policies.
• Residual conditioning significantly improves performance, reducing directional Shape RMSE from 0.748 to 0.555 mm with diffusion policies.
• Voxelized point-cloud Explicit policy with residuals achieves 0.534 mm depth and 0.443 mm directional control accuracy, generalizing well to unseen targets and biological tissue.