Smoothness Constrained Curiosity Driven Multicamera Trajectory Optimization for Robot-Assisted Minimally Invasive Surgery

• Main Authors: Subedi, Divas, Jiang, Wenfan, Rahman, Ramisa Tahsin, Zhang, Heidi, Huang, Kevin, Su, Yun-Hsuan
• Format: Conference Proceeding
• Language: English
• Subjects: 3D reconstruction; Laparoscopes; Measurement; Medical robotics; Minimally invasive surgery; multirobot systems; path smoothness; reinforcement learning; Robot vision systems; robot-assisted surgery; Surface reconstruction; Three-dimensional displays
• ISSN: 2771-9049
• DOI: 10.1109/ISMR57123.2023.10130237

This paper presents a novel, curiosity driven camera positioning algorithm for multicamera systems in robot-assisted minimally invasive laparoscopic procedures. The work here extends the authors' prior studies in curiosity driven movement algorithms by introducing a new jerk-minimization term in the extrinsic curiosity reward function. Three basic and three curiosity driven movement baseline algorithms are comparatively evaluated against the novel motion-smoothing approach in both visual and motion metrics - the latter is analyzed with both time and frequency domain scores. All tests were performed on an identical laparoscopic simulation, with dynamic tissue surface, simulated breathing motion, and virtual tool-tissue interactions incorporated. Multicamera systems can be inserted through a single trocar and be magnetically anchored or maneuvered. Such systems can enable enhanced visual feedback and sensing of the surgical cavity, with multiple simultaneous views affording 3D reconstruction. Results of the study presented here are promising, and show that the modified curiosity driven algorithm does indeed reduce camera jerk at a meager cost to visual and reconstructability metrics. It is hypothesized that reduction in overall camera motion jerk, while still maintaining 3D reconstructability quality, is a desirable characteristic in teleoperated laparoscopic procedures.