Automated Trajectory Generation for Robotic Surgical Tasks

Robots nowadays have been widely adopted in Minimally Invasive Surgery (MIS), and task autonomy is one of the key steps towards autonomous robotic surgery. In surgical tasks, the movements of every surgical instrument require high precision and accuracy with safety considerations, since abrupt movements may damage patients’ tissues or even lead to life-threatening conditions. With the aim of surgical task autonomy, the majority of the approaches depend on the use of keypoints where keypoints are usually defined by surgeons first, and the instrument movement trajectory could then be planned based on these keypoints. In this paper, Surgical Trajectory generation with VAE-Transformer (SurgTraj-VAETrans), a neural network model, is proposed to automatically generate trajectories for robotic surgical tasks by learning from the limited number of demonstrations. These demonstrations are performed by participants using da Vinci Research Kit (dVRK) on a specific surgical task. The results are evaluated by predefined metrics and compared with some well-known generative models. The result shows that the proposed model can generate reliable trajectories for surgical task purposes.