Using data estimation to improve performance of the da Vinci Surgery Robot on autonomous tasks

Simulation:

SOFA will be used to include a da Vinci research kit 3D model that is open to the public in URDF and stl files, then similar deformable tissues analysis similar to those used in Tagliabue, E., Dall’Alba, D., Pfeiffer, M. et al. "Data-driven Intra-operative Estimation of Anatomical Attachments for Autonomous Tissue Dissection" IEEE Robotics and Automation Letters (2021), will be used to detect the deformable properties of human tissues. 

In addition, manipulation task such as handling tissues will be performed using SOFA deformable object capabilities with pytorch and a Q-learning model that tries to minize the distance between the point of contact of the PSM and the tissues. For the cutting procedure, an additional PSM arm will perform a cut on the middle of a tissue following a Q learning method that is based on geometric constrains. In the case of suturing, the PSM arm will find fix points in the tissue that will be sutured and will follow these geometrical constrains.

Data gathering:

-For all tasks, kinematics and dynamic data will be colected as well as the machine learning performance. The same experiments will be recreated on a real da Vinci Research Kit that is located in the lab at the University Texas at Austin.

-Performance metrics:

The real2sim accuracy will be obtained and measured from the simulation data and the real experiments.