http://robotic.media.mit.edu/projects/robots/tikl/tikl.html
When learning a motor skill, such as when in rehabilitation or dance class, people learn through several means: through sight, sound, and touch. The most difficult method of instruction is through touch, since the instructor also needs to be performing the action and also cannot guide all joints of the student. This research aims to create a suit that is embedded with vibrotactile actuators at each joint to give corrective feedback to help teach the motor skill to the student.
The idea is that the instructor will wear a suit with motion sensors and will be tracked with a tracking system. The student also wears a suit equipped with motion sensors and vibrotactile actuators. When the student mimic's the instructor's movements, the suit will vibrate the joints that are incorrect proportional to the error in posture.
The suit they have created covers the wrist, elbow, shoulder, and chest. Motion sensors are used and tracked by a Vicon system to model the subject's arm. The vibrotactile actuators are placed around the wrist and elbow joints. Four actuators are placed around each joint. When the control software detects that the subject is not positioned correctly, the actuators vibrate in a specific way. For rotational correction, the actuators vibrate clockwise or counterclockwise around the joints to simulate torque on the joint. For joint angle correction, two of the sensors (top and bottom of the wrist) vibrate. One of them will vibrate with more intensity, indicating the direction the hand must move. It is intended that there is a vibrational "force field" around the proper join angles.
The suit is connected to a computer, and control software processes the motion capture data from the teacher and the student (the user) to determine the student's errors. The error data is sent to a custom hardware controller which transforms the error data into the proper vibrations.
A user study was performed with 40 subjects. Half the subjects received visual correction while the other 20 received visual correction plus vibrotactile feedback. Each person in each group was shown the same videos of an instructor posing in a specific position. The videos were shown in a random order and were only a few seconds long to ensure that no subject would be able to perfectly mimic every pose.
In a questionnaire given immediately after the test, subjects who had the vibrotactile feedback felt that it didn't help much at the beginning but might help over time if they kept using the system. The vibrotactile group also reported more fatigue than the non-vibrotactile group.
The studies were video recorded, and analysis of the video shows enhanced performance with vibrotactile feedback. The error rates computed by the control software show consistently lower errors in all trials.
Hinge joint corrections were found to be easier to understand than rotational joint corrections. The error rates for hinge joint correction are much lower for the vibrotactile feedback group, while error rates for rotational joint correction are not significantly different for the vibrotactile feedback group.
Overall, the study shows a significant gain in performance (up to 27%) and accelerated learning (up to 23%). It is pointed out that the positions of the vibrotactile actuators and the methods of feedback were not optimized, so it is expected that these number will improve even more as the technology is optimized and a full-body suit is created.
For future work on the technology itself, Lieberman Breazeal plan on finding a lower cost relative positioning system to aid in the adoption of this technology. They also want to develop or find smaller, more powerful tactile actuators. They want to increase the performance and learning gains with rotational joints. They wonder if this system can be successfully scaled up to the full human body, since there would be around 100 actuators and the body might not be able to handle that. They want to learn the long-term effects of this system as well.
In other future work, they want to test this system with less or no visual feedback, so this can be used for visually impaired or blind people. They also want to explore neurological rehabilitation and posture correction.
------------
I thought this work was very interesting and presents a great use of vibrotactile technology. I have been thinking of other ways to use this technology than the project Manoj and I are currently working on with our vibrotactile gloves. I especially liked the possibility of using the suit to help blind people perform certain tasks or learn certain motor skills, since that is related to my own work, which is to help blind people see drawings and images and also draw using vibrotactile feedback to the fingers.
I was impressed that the system worked well in its initial unoptimized, simplified form (consisting of just one arm). I hope this suit will perform on par with or better than this initial study.
I was wondering why they used video and images of the instructor in the user study rather than a live instructor. I would be interested to see how the suit performs with a human instructor. This would be a different form of visual stimulus, and might be superior.
2 comments of glory:
I also found the paper fascinating, no limit to the applications. You mentioned working with the blind, I suspect some of these tasks with feedback of this nature may be more readily acquired in the dark or with ones eyes closed after an initial familiarity has been established.
about the live instructor, I agree it would be a much better user study. however, to make everything fair, the instructor would have to do the exact same thing for every test subject to make it is a fair experiment. that would be difficult to reproduce the exact same motion 40 times.
Post a Comment