A Fundamental Theory for Dexterous Surgical Skills Transfer to Medical Robots
|Interdisciplinary Areas:||Engineering and Healthcare/Medicine/Biology, Data/Information/Computation, Human-Machine/Computer Interaction, Human Factors, Human-Centered Design
When tele-operation is challenged due to limited bandwidth, latency and lost-of-signal, autonomy needs to step in. While modern robots have been endowed with excellent sensing and dexterous capabilities, there is a gap in knowledge about how to “transfer” automatically existing abundant knowledge about surgical maneuvers from the operating room (OR) to new, uncontrolled and austere settings. The goal of this research, is to study and theorize about new approaches to predict information that will determine at what extent a robot needs to autonomously complete surgical procedures, making use of a sequence of maneuvers associated with the required surgical procedure extracted from a previously learned libraries (high level encoded knowledge). To populate this library, there is a new to theorize new approaches for transfer learning, where learnt patterns will be projected to fundamentally different domains with variable resource constraints (transfer learning).
The postdoctoral research objective is to develop theoretical frameworks for supervised autonomy, capable to self-adjust its autonomous behavior and perform procedures in never seen settings using a transfer learning paradigm. The working hypothesis is that an existing surgical procedure can be adapted to a new domain using an encoding scheme to restore supervisory content combined with a one shot learning framework.
The successful candidate will be able to derive theoretical solutions to components that contribute to the above grant, implement prototype systems that demonstrate the theory, and work with an existing team to integrate and prove that code within a large system.
Superior system development experience (including programming and implementation of Deep Learning and Computer Vision techniques; (MATLAB experience alone is insufficient) is required and should be evident in the candidate's CV.
Preferred qualifications include: * PhD needs to be recent, finalized before start date * preferred PhD discipline is Electrical and Computer Engineering and/or Computer Science; a classical, broad, CS education is important * strong publication record at top robotics and computer vision venues (e.g., ICRA, IROS, RSS, CVPR, ICCV, ECCV, BMVC, IEEE PAMI) * experience with a variety of methodologies (machine learning is important but even more so is the full spectrum of robotics and computer vision methods) * documented interests and experience in robotics and deep learning driven by medicine. This is a basic research position, but with connections to medical robotics, so evidence of basic research accomplishment is important.
Juan P. Wachs
1. Learning by Observation for Surgical Subtasks: Multilateral Cutting of 3D Viscoelastic and 2D Orthotropic Tissue Phantoms
2. Do what I want, not what I did: Imitation of skills by planning sequences of actions
3. System events: readily accessible features for surgical phase detection
4. Autonomous Multilateral Debridement with the Raven Surgical Robot
5. Recognizing Surgical Activities with Recurrent Neural Networks