Will Panitch

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2024-127

May 17, 2024

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-127.pdf

The introduction of robot surgical assistants (RSAs) like Intuitive Surgical’s da Vinci system has equipped surgeons with an additional set of powerful tools for constrained, precise, and endoscopic manipulation. These robots enable medical professionals to perform an array of previously impossible minimally-invasive procedures that result in better medical outcomes, less scarring, and faster recovery. Additionally, RSAs have the potential to standardize procedures and reduce surgeon fatigue through supervised subtask automation. Certain oft-performed, repetitive subtasks, such as incision closure and debridement, could be autonomously performed under surgeon supervision, eliminating certain time-consuming and tedious tasks from the surgeon’s workload. To advance this line of research, we propose a unified toolkit for surgical augmented dexterity, consisting of a U-Net-based visual localization module that is trained using autonomously collected subtask data, as well as adaptations of the aforementioned model for 3- or 6-D localization of different common surgical objects, and a set of learned servoing modules that enable critical fine motor control tasks in the surgical setting, even under unreliable proprioceptive feedback. We then apply this sensing-and-planning paradigm to two common surgical subtasks: suturing and vascular shunt insertion, and demonstrate that it enables state-of-the-art autonomous task performance. The augmented dexterity framework achieves an average of 2.93 consecutive completed suture throws using unmodified surgical grippers and needles (important for ensuring instrument sterility), and demonstrates a 75%–100% success rate on different vessel phantoms in the shunt insertion task. These results validate the utility of the framework, and help demonstrate a potential path towards increasing subtask autonomy for surgical settings.

Advisors: Ken Goldberg

---

BibTeX citation:

@mastersthesis{Panitch:EECS-2024-127,
    Author= {Panitch, Will},
    Title= {Toward Autonomous Endoscopic Surgery: a Framework and Case Studies for Robotic Learning in Healthcare},
    School= {EECS Department, University of California, Berkeley},
    Year= {2024},
    Month= {May},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-127.html},
    Number= {UCB/EECS-2024-127},
    Abstract= {The introduction of robot surgical assistants (RSAs) like Intuitive Surgical’s da Vinci system has equipped surgeons with an additional set of powerful tools for constrained, precise, and endoscopic manipulation. These robots enable medical professionals to perform an array of previously impossible minimally-invasive procedures that result in better medical outcomes, less scarring, and faster recovery. Additionally, RSAs have the potential to standardize procedures and reduce surgeon fatigue through supervised subtask automation. Certain oft-performed, repetitive subtasks, such as incision closure and debridement, could be autonomously performed under surgeon supervision, eliminating certain time-consuming and tedious tasks from the surgeon’s workload. To advance this line of research, we propose a unified toolkit for surgical augmented dexterity, consisting of a U-Net-based visual localization module that is trained using autonomously collected subtask data, as well as adaptations of the aforementioned model for 3- or 6-D localization of different common surgical objects, and a set of learned servoing modules that enable critical fine motor control tasks in the surgical setting, even under unreliable proprioceptive feedback. We then apply this sensing-and-planning paradigm to two common surgical subtasks: suturing and vascular shunt insertion, and demonstrate that it enables state-of-the-art autonomous task performance. The augmented dexterity framework achieves an average of 2.93 consecutive completed suture throws using unmodified surgical grippers and needles (important for ensuring instrument sterility), and demonstrates a 75%–100% success rate on different vessel phantoms in the shunt insertion task. These results validate the utility of the framework, and help demonstrate a potential path towards increasing subtask autonomy for surgical settings.},
}

EndNote citation:

%0 Thesis
%A Panitch, Will 
%T Toward Autonomous Endoscopic Surgery: a Framework and Case Studies for Robotic Learning in Healthcare
%I EECS Department, University of California, Berkeley
%D 2024
%8 May 17
%@ UCB/EECS-2024-127
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-127.html
%F Panitch:EECS-2024-127