- 89 views
- 85 downloads
Motion Planning of Robotic Systems in Diagnostic and Therapy Applications Using Control and AI
-
- Author / Creator
- Akbari, Mojtaba
-
This thesis presents significant research on robotic motion planning within diagnostic and therapy applications, with a primary focus on the integration of control and AI techniques. The research encompasses three main contributions: a robotic ultrasound imaging method, a robot-assisted ultrasound scanning system, and an innovative framework for uncertainty-aware control in medical robots.
The first component of this thesis introduces a robotic ultrasound imaging method that employs a five Degrees of Freedom (DoFs) robotic system. The primary objective is to achieve precise scanning of breast tissue for the generation of high-quality ultrasound images. This method initiates with a pre-scan phase, wherein geometric analysis of the target within the breast is utilized to determine the desired scanning trajectory. Subsequently, in the post-scan phase, the probe's rotational and translational movements are continually adjusted based on the center of mass of segmented targets within each acquired frame and the average confidence map of the images. The experimental validation of this visual servoing algorithm on a plastisol phantom demonstrates the system's proficiency in controlling the ultrasound probe's motion, effectively targeting tissue, and efficiently executing real-time robotic control loops.
The second significant contribution of this thesis is the development of a robot-assisted ultrasound scanning system, a response to the challenges posed by the COVID-19 pandemic. Traditional ultrasound scans require close contact between the sonographer and the patient, which increases the risk of disease transmission. This novel system mitigates this risk by automating tissue scanning through a dexterous robot arm holding the ultrasound probe. The system constantly evaluates the quality of acquired ultrasound images in real-time using a quality assessment algorithm based on correlation, compression, and noise characteristics. Feedback from the ultrasound images guides the system in automatically adjusting the probe's contact force, ensuring consistent high image quality. An SVM classifier analyzes image features and provides feedback to the robot arm for precise force adjustments. Experimental trials conducted on plastisol phantom tissue confirm the system's capacity to maintain image quality while minimizing direct sonographer-patient contact.
The third and equally crucial contribution of this thesis centers around addressing safety concerns and uncertainty analysis in deep learning-based medical robotic applications for motion planning. The integration of deep learning algorithms into medical robots introduces uncertainties that can compromise the safety of both patients and the overall operation. To tackle this challenge, a pioneering framework for uncertainty-aware control of medical robots is introduced. This framework is particularly designed for a lower-limb exoskeleton intended to assist individuals with disabilities. The framework leverages fast uncertainty analysis within the medical robot's control loop. By quantifying uncertainty levels during both training and testing phases, the proposed framework ensures safe and reliable human-robot interactions.
During the training phase, the framework employs Kullback-Leibler (KL) divergence to identify similarities between labels and predictions. In the testing phase, it utilizes Mahalanobis distance to detect out-of-distribution (OOD) data, enhancing safety and improving decision-making for the robot controller. Experimental trials conducted on the ExoH3 lower-limb exoskeleton illustrate the effectiveness of this uncertainty analysis technique in real-time motion planning and its ability to identify OOD features that may lead to unsafe motion execution.
In summary, this thesis represents a significant advancement in the field of motion planning for robotic systems in diagnostic and therapy applications, particularly addressing critical challenges related to safety and uncertainty. The proposed approaches for robotic ultrasound imaging, robot-assisted ultrasound scanning, and uncertainty-aware control have the potential to enhance the efficacy and safety of medical robotics, ultimately benefiting both patients and healthcare providers.
-
- Graduation date
- Fall 2023
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.