Robotics Research Lab
CRES
USC Computer Science
USC Engineering
USC
/ Research / Projects / A 'Hands Off' Physical Therapy Assistance Robot

Overview Approach Evaluation Images Videos Publications Contact Details Support

Overview

We seek to empirically study the effect of different types of interaction between a robotic assistant and a cardiac surgery patient. We hope to compel the patient to reach their physical therapy goals by providing engaging feedback and encouragement. The effect of different feedback techniques will be compared through a controlled experiment with patients at USC University Hospital. Patients of cardiothoracic surgery are required to use special breathing equipment every hour in order to prevent pneumonia and speed recovery. Nurses are too busy to oversee all the sessions throughout the patient's stay, and patients quickly lose their internal motivation to complete the exercises because of physical discomfort or boredom. We expect that the robot will be able to help patients complete their excercises, leading to a faster recovery.

Using interaction without manipulation is a unique kind of rehabilitation robotics. The overwhelming majority of current and past rehab robotics projects have involved manipulation of some kind. There are many legal liabilities and technical difficulties when robots physically contact patients. A mobile robot that simply interacts with patients avoids these problems, and can therefore be cheaper and more practical for widespread home use.

Approach

The robot uses a laser rangefinder to safely navigate within the patient's hospital room. The patient is located visually using a face-finding algorithms on the robot's stationary webcam. The patient will be able to communicate with the robot through head nodding and shaking. A behavior will monitor optical flow in the region of the patient's face to detect these cues. The patient should be able to use head shaking to interrupt the robot. This is useful for the patient to indicate when they no longer are able to proceed with the excercises in a natural way.

The interaction centers around the robot encouraging the patient to perform deep inhalations using a spirometer. This is a small, inexpensive device that gives a visual indication of air intake in a given breath. The robot can track the spirometer by searching for a simple pattern of colored ficucial patches. A small square of pink reflective cardboard is fixed to the lower corner of the device so the detector can search for the pink blob with an equal sized yellow patch directly above it. The occurrence of this type of pattern will be infrequent enough in the hospital room environment to allow robust tracking.

spirometer

The location of the device will be used to bootstrap tracking of the volume indicator bar. Its height indicates the current level of air that has been taken in by the patient. Another colored marking affixed to the bar will aid in tracking.

Verbal interaction will begin after the robot introduces itself, and asks the patient whether they need to be reminded how to use the spirometer device. The form of recurring questions like this one will vary across the different sessions according to a dynamic script. This way the robot will first ask: "Do you need instructions for using the spirometer", but in later sessions it will vary its question to something like: "Do you need to be reminded about how to use the spirometer", his will provide a more natural interaction.

Evaluation

The main variable in the robot's interaction will be the way it attempts to reward patients for reaching a goal number of inhalations. Several methods will be investigated including verbal praise, celebratory music, flashing lights, and entertaining motion routines such as rotating pinwheels.

Evaluation of the robot will be determined primarily by whether the patients exhibit clinical improvement with respect to a control group. Also, the patient's own experience of the robot can serve as a second criteria for evaluating performance. Tradeoffs may exist between these two measures, but we hope to be able to maximize clinical improvement while still providing a more pleasureable experience for patients.

Images

Clara is the assistive robot of the Interaction Lab. The robot was named after Clara Barton, who founded the American Red Cross.

A prototype of Clara (front)
A prototype of Clara (side)
Clara with clothes
Clara interacting with a test subject
ICORR-05 poster

Videos

  • Simple demonstration with flash cards. [AVI, 18MB]
  • Experiment 1 with a test subject conducted at the Interaction Lab. [MOV, 4MB]
  • Experiment 2 with a test subject conducted at the USC University Hospital. The speech recognition system was deactivated due to the noizy environment. [MOV, 2.4MB]
  • Publications
    1. Kyong Il Kang, Sanford Freedman, Maja J. Mataric´, Mark J. Cunningham, and Becky Lopez. "A Hands-Off Physical Therapy Assistance Robot for Cardiac Patients". In Proceedings, International Conference on Rehabilitation Robotics, Chicago, Illinois, Jun 28-Jul 1 2005.(.pdf)
    2. Kyong Il Kang, Sanford Freedman, Maja J. Mataric´, Mark J. Cunningham, and Becky Lopez. "A Hands-Off Physical Therapy Assistance Robot for Cardiac Patients". CRES Technical Report CRES-05-001, 2005.(.pdf)
    Contact Details

    Kyong Il Kang
    Sanford Freedman

    Support

    This work is supported by USC Provost's Center for Interdisciplinary Research (CIR) Fellowship and the Okawa Foundation.