Supervised Mixture-of-Experts for Surgical Grasping and Retraction

TL;DR

This paper introduces a supervised Mixture-of-Experts architecture for surgical manipulation tasks, enabling effective learning from limited data and demonstrating robust generalization and transfer capabilities in robotic surgery scenarios.

Contribution

The paper presents a novel supervised MoE architecture that enhances surgical robot learning, allowing complex manipulation from fewer demonstrations and improving robustness and transferability.

Findings

Supervised MoE significantly improves success rates over standard models.

The approach generalizes to unseen viewpoints and transfers zero-shot to ex vivo tissue.

It demonstrates promising preliminary results in in vivo porcine surgery.

Abstract

Imitation learning has achieved remarkable success in robotic manipulation, yet its application to surgical robotics remains challenging due to data scarcity, constrained workspaces, and the need for an exceptional level of safety and predictability. We present a supervised Mixture-of-Experts (MoE) architecture designed for phase-structured surgical manipulation tasks, which can be added on top of any autonomous policy. Unlike prior surgical robot learning approaches that rely on multi-camera setups or thousands of demonstrations, we show that a lightweight action decoder policy like Action Chunking Transformer (ACT) can learn complex, long-horizon manipulation from less than 150 demonstrations using solely stereo endoscopic images, when equipped with our architecture. We evaluate our approach on the collaborative surgical task of bowel grasping and retraction, where a robot assistant interprets visual cues from a human surgeon, executes targeted grasping on deformable tissue, and performs sustained retraction. Our results show that generalist Vision Language Action models fail to acquire the task entirely, even under standard in-distribution conditions. Furthermore, while standard ACT achieves moderate success in-distribution, adopting a supervised MoE architecture significantly boosts its performance, yielding higher success rates in-distribution and demonstrating superior robustness in out-of-distribution scenarios, including novel grasp locations, reduced illumination, and partial occlusions. Notably, it generalizes to unseen testing viewpoints and also transfers zero-shot to ex vivo porcine tissue without additional training, offering a promising pathway toward in vivo deployment. To support this statement, we present qualitative preliminary results of policy roll-outs during in vivo porcine surgery.