Error Compensation in Human-machine Interfaces

Inherent in normal human hand motion are numerous position error components. These include physiological hand tremor, jerk, and low-frequency wander. In addition, pathological movement disorders, stemming from disease or injury, introduce further position error due to conditions such as pathological tremor, athetosis, and ataxia. These disorders are often severe.

 In patients with movement disorders, involuntary motion interferes with quality of life and independence in daily living. In microsurgery, involuntary hand motion hampers performance, particularly in the ophthalmological and neurological fields.

 My primary research interest is in error compensation in human-machine interfaces. Applications include:

My work to improve manual precision in human-machine interfaces has led me to research in three areas:

1. Characterization of human motion

This work has involved measurement of tremor and non-tremor errors in motion in disabled and healthy individuals. Most recently, I have completed a study quantifying low-frequency position errors in microsurgery. Publications stemming from this work include:
 
 

2. Algorithms for error compensation

Adaptive tremor canceling

This continuing work began as my dissertation research at Johns Hopkins University.

There are two types of tremor: physiological tremor, which is present in all human motion, and pathological tremor, caused by injury or diseases such as essential tremor, Parkinson's disease, and multiple sclerosis. Pathological tremor greatly degrades manual control of motion. Physiological tremor causes imprecision in fine motor tasks such as microsurgery.

I developed a new adaptive filter to perform adaptive noise canceling of tremor in human-machine interfaces. I demonstrated its use in three practical applications:

For more information see the online publications (and source code) listed below, or the Publications page.

Augmentation of manual precision

This work may be thought of as a superset of the adaptive tremor canceling work. Tremor is not the only source of manual position error. Hand motion during manipulation contains considerable low-frequency error, or drift. Suppressing this error is difficult because it overlaps in frequency with voluntary motion. Furthermore, little is known about the nature of this component of error, or its origin within the human system.

 Because so little is known about I am applying cascade neural networks to learn overall patterns of instrument position error during microsurgery. These neural networks can then be used online to estimate the error in hand motion, and send a drive signal to actuators in the tip of an active handheld microsurgical instrument to compensate for the error in real time.

3. Electromechanical hardware for error compensation

Relevant publications include:
 
 

Publications available online

Source code available online

Carnegie Mellon Computer ScienceCam.Riviere@cmu.edu