|
|
Foundations of Robotics Seminar, May 2, 2007
Time and Place | Seminar
Abstract | Speaker Appointments
A Neuromuscular Framework for Motor Control
Pedram Afshar
NSH 1507
Refreshments 4:15 pm
Talk 4:30 pm
The emergence of brain-controlled interfaces (BCI) has
provided a novel avenue to help restore function for patients with paralysis
and peripheral neuropathy. The proposed mechanism for many BCI’s
involves decoding a patient’s desired motor output from neural signals
and recreating it using a kinetic effector (for example, robotic prosthetics or
functional electrical stimulation), thereby bypassing the damaged parts of the neuromusculoskeletal system. Like biological limbs, these effectors are
controlled like a marionette: joints are not moved directly, but by actuating
cables (or tendons) attached to the links (or bones). Thus, controlling a kinetic effector for BCI
or a human limb ultimately requires understanding the relationship between
actuator control inputs and the motor output.
This thesis describes a framework for describing the
relationship between the control inputs and the resulting motor output for both
kinetic effectors and biological limbs. The
framework unifies previously disparate concepts of motor output (including limb
kinematics, endpoint force, and endpoint stiffness) and allows mechanical
constraints to be translated into the control input and motor output domains. This framework was applied to the index
finger, a limb central to human interaction with the environment. The framework captured the biomechanical
constraints of the index finger to reveal structure in the motor output and
provide a context for its control input.
Structure in the motor output showed that most of the variability in
index finger stiffness could be explained using a single scalar, contrary to
the traditional description of limb stiffness requiring 6 scalar values. In addition, human index finger control
strategies were shown to have much less variance than available from the context
of theoretically possible control inputs.
This finding implies a potential role for neural constraints or
optimizations in the selection of index finger control strategies. Together, these findings imply that finger
control is much simpler than predicted based only on its kinematic
complexity.
In the future, applications of this framework can be
used not only to direct the fabrication and control of kinetic effectors, but
also as a tool to investigate the neural strategies used for human limb control.
For
appointments, please contact Pedram Afshar.
The Robotics
Institute is part of the School of Computer
Science, Carnegie Mellon University.