Latent regression based model predictive control for tissue triangulation
Songtao Liu, Jacinto Colan, Yaonan Zhu, Taisuke Kobayashi, Kazunari Misawa, Masaru Takeuchi, Yasuhisa Hasegawa
Abstract
Tissue triangulation is a fundamental surgical skill in laparoscopic surgery that requires the simultaneous control of multiple surgical tools to create tension and enable tissue visualization. Robotic assistants can facilitate triangulation tasks by autonomously operating some of the tools, reducing the surgeon's cognitive workload and improving precision and safety. However, characterizing complex tissue dynamics is a significant challenge for robot-assisted tissue triangulation. This work presents Latent Regression based Model Predictive Control (LR-MPC) for tissue triangulation that guides the tissue state to lie within a target distribution in the secondary latent space. A Model Predictive Control strategy in latent space is followed for optimal robot action selection. We evaluated the proposed method in a robot-assisted triangulation task with various sizes and materials of triangular tissues. The results show that LR-MPC can achieve performance comparable to a human assistant's, demonstrating its effectiveness and robustness for autonomous tissue triangulation.