Abstract
The outbreak of COVID-19 has led to a shortage of medical personnel and an increasing need for nucleic acid testing. Manual oropharyngeal sampling is susceptible to inconsistency caused by fatigue, and close contact can expose healthcare personnel to cross-infection. This paper proposes a fully autonomous oropharyngeal-swab robot system installed in a negative pressure chamber. The system carries out a standardized sampling process to minimize individual differences. A hierarchical throat detection algorithm is presented, fusing multi-modal sensory information to localize the optimum sampling location accurately. Furthermore, a force/position hybrid control method ensures both accurate sampling and subject comfort. Preliminary clinical trials show that the robot system can safely and efficiently collect samples, providing a scalable solution for large-scale PCR molecular sample collection for various respiratory diseases.
Introduction
During the COVID-19 pandemic, manual throat swabbing became a significant bottleneck in public health efforts. The traditional manual process suffers from three major flaws: the risk of healthcare worker infection through aerosol exposure, inconsistency in sample quality due to human fatigue, and extreme labor intensity. While semi-automatic or teleoperated robots have been explored, they still require significant human oversight. Researchers from Tsinghua University have addressed these challenges by developing the first fully autonomous robot system capable of performing oropharyngeal swabs without human intervention.
System Architecture and Workflow
The system architecture is split into two main robotic components: a 6-DOF loading arm (JAKA ZU3) for handling swabs and storage tubes, and a custom 3-axis sampling robot designed for the delicate task of swabbing. The entire system is housed in a negative pressure chamber to prevent the spread of viral aerosols.
As illustrated in the system topology, the process begins with a vocal interaction module that guides the subject into position. The system then executes a synchronized workflow where the loading robot prepares the swab while the sampling robot, guided by an RGBD camera and tactile sensors, identifies the target and performs the collection.
Technical Innovations in Detection and Control
The core of the system's autonomy lies in its hierarchical computer vision pipeline and advanced control strategy:
- Hierarchical Throat Detection: The robot uses a "coarse-to-fine" vision pipeline. It first detects the face and identifies 68 facial landmarks to locate the mouth. Finally, a self-aligning multi-modal transformer pinpoints the exact swabbing location on the posterior pharyngeal wall by fusing RGB, Depth, and Infrared (IR) data.
- Force/Position Hybrid Control: To ensure subject safety, the robot utilizes a tactile sensor at the tip of the swab. If the interaction force exceeds 1 Newton, the system immediately halts or retracts, with a rapid response time of 20ms.
- Optimized Localization: The researchers utilized "Wing Loss" for their localization model, which is more sensitive to small errors than traditional L1 or L2 loss functions—a critical factor when targeting the small, deep area of the throat.
Standardized Sampling Strategy
To ensure sample validity, the robot does not simply touch the target area; it mimics the sophisticated motion of human medical staff.
The system performs a programmed 180-degree repetitive arc across the posterior pharyngeal wall. This standardized wiping motion ensures the collection of sufficient epithelial cells for a valid PCR test, eliminating the variability inherent in manual collection.
Key Results and Performance
Clinical evaluations and technical testing have validated the system's efficacy:
- Precision: The mean sampling point deviation was recorded at 3.72 mm, well within the effective swabbing area of the human throat (approx. 160 $mm^2$).
- Efficiency: The complete cycle—from the subject sitting down to sample storage—takes approximately 166 seconds, allowing for a throughput of roughly 20 subjects per hour.
- Safety and Validity: All tactile forces remained within safe thresholds (below 1N) during trials. RT-PCR tests confirmed that the robotically collected samples maintained high quality, with Ct values consistently within the qualified range for diagnosis.
Conclusion and Future Work
This autonomous robot system offers a safer and more consistent alternative to manual swabbing, reducing medical worker exposure and alleviating labor pressure. While the current throughput is limited by the mechanical loading and unloading of swabs, future iterations will focus on high-speed loading devices to increase capacity. Additionally, further development will aim to improve the system's robustness when interacting with non-compliant subjects or those with unique anatomical challenges.
Summary updated