Sterility-Aware Robot Learning
Bedside nursing procedures like central line dressing changes require not just task competence, but strict sterility throughout. This NIH R21 proposal builds a computational framework that learns clinical sterility constraints directly from 380 expert nurse demonstration videos — then uses those learned constraints to guide real-time, clinically compliant robot execution.
The Problem
Sterility violations during bedside nursing procedures directly increase patient infection risk — yet existing robot learning pipelines treat sterility as an afterthought. A robot that completes the task but breaks sterile technique doesn't just fail: it actively harms the patient. This proposal asks whether sterility constraints can be learned from expert nurse demonstrations and used to guide real-time robot execution, rather than hand-engineered after the fact.
Research Approach
The project has three interconnected components. First, we build a perception pipeline that learns sterility-relevant features — object positions, contamination zones, safe handling regions — from annotated nurse demonstration videos. Second, those features train continuous constraint models that score any robot configuration as admissible or at risk. Third, the learned constraints are integrated directly into the robot's motion planner, so the system enforces sterile technique in real time during physical execution.
My Role
I plan to contribute the tactile sensing layer — safe clinical grasping requires the robot to feel, not just position. Tactile sensing on the end-effector can provide the contact and proximity feedback needed for compliant grasps that don't break the sterile field. The specific sensing implementation is still being worked out. I'll also be helping with data collection and robot integration.
Team
Principal Investigator — University of Houston
Pam Qian
Leads the proposal and contributes the nurse demonstration dataset collected at Houston Methodist Hospital, bringing expertise in clinical robotics and human-robot collaboration in healthcare settings.
Co-Investigator — University of Colorado Boulder · Correll Lab
William Xie
Co-investigator on the proposal, contributing expertise in tactile sensing and robot learning from the Correll Lab at CU Boulder.
Co-Investigator — University of Colorado Boulder
Nikolaus Correll
Contributes robot sensing and manipulation expertise from the Correll Lab at CU Boulder, supporting the tactile integration and robot execution components of the proposal.
Researcher — University of Colorado Boulder · Correll Lab / HIRO Lab
Artemis Shaw
Planned contributions include tactile sensing for compliant clinical grasping, data collection, and robot integration. The specific sensing implementation is still being determined — work is contingent on proposal funding.