Error Analysis and Control for the Patient Positioning System

MGH Research Project

 

Sponsor

National Institutes of Health (NIH), via the Massachusetts General Hospital (MGH)

Principal Investigator

 Professor S. Dubowsky

Group Members

Dr. Steven Dubowsky

Constantinos Mavroidis, PhD

Marco Meggiolaro, PhD student

Phillipe Drouet, PhD

 

Problem Statement

This page describes the NIH sponsored research conducted in the Field and Space Robotics Laboratory at MIT, via the Northeast Proton Therapy Center from the Massachusetts General Hospital.

The motivation of this research comes from the need of a high accuracy robotic patient positioning system in a radiation therapy research facility now being constructed at the Massachusetts General Hospital (MGH), the Northeast Proton Therapy Center (NPTC). The robotic patient positioning system (PPS) places a patient in a high energy proton beam delivered from a rotating gantry structure. The PPS is a six degree of freedom manipulator that covers a large workspace of more than 4m in radius while carrying patients weighing as much as 300 lbs. Patients are finally immobilized on the "couch" attached to the PPS end-effector. The PPS, combined with the rotating gantry that carries the proton beam, enables the beam to enter the patient from any direction, while avoiding the gantry structure. Hence programmable flexibility offered by robotic technology is needed.

The required absolute positioning accuracy of the PPS is ±0.5 mm. This accuracy is critical as larger errors may be dangerous to the patient. The required accuracy is roughly 10-4 of the nominal dimension of system workspace. This is a greater relative accuracy than many industrial manipulators. In addition, FEM studies and experimental results show that the changing and heavy payload (between 1 and 300 pounds) creates end-effector errors due to elastic deformations of the order of 6-8 mm.

Research

Due to task constraints it is often not possible to use direct end-effector sensing in a closed-loop control scheme to improve the system accuracy. Therefore, there is a need for model based error identification and compensation techniques. While classical calibration methods can achieve such compensation for some systems, they cannot correct the errors in large systems with significant elastic deformations, because they do not explicitly consider the effects of task forces and structural compliance. A method has been developed that considers both deformation and more classical geometric errors in a unified manner.

The method explicitly considers the weight dependency of the errors. An error model and a set of experimentally measured positions and orientations of the robot end-effector, and measurements of the payload wrench, are used to calculate the robot "generalized" errors without needing a manipulator elastic model. Generalized are called the errors that characterize the relative position and orientation of frames defined at the manipulator links. They are found from measured data as a function of the configuration of the system and the task forces. Knowing these generalized errors the manipulator end-effector position and orientation errors are calculated and used at any configuration to correct the robot configuration to compensate for these errors.

The method treats all errors of the manipulator such as geometric and elastic errors in a unified manner. The method was applied to the Patient Positioning System. In this work, 450 measurements were used to evaluate the basic accuracy of the PPS, and later used to evaluate the accuracy of the compensation method. A force/torque sensor has been added to the system to measure the wrench applied by the patient’s weight. It was experimentally shown to be able to reduce the inherent 5-7mm errors to less than the required accuracy of 0.5 mm.

Papers

Identification And Compensation Of Geometric And Elastic Errors In Large Manipulators: Application To A High Accuracy Medical Robot

Pictures

Schematic of the PPS and the Gantry

 

The Patient Positioning System

 

Copyright 1995-2001 by MIT.

All rights reserved.