mation to the deaf and blind community through the skin.
Similar work has shown scaled architectures of almost 1000
DOFs, for video tactile displays. However, touch has not yet
been used to augment our somatosensory learning systems.
Promising work has hinted at the utility of tactile feedback
for neurological trauma rehabilitation. But as of yet, research
has focused on the application of torques on subjects’ joints.
Patients in neurological literature (such as HM [12]) have
lost the ability to form new long-term memories, but can still
build new motor skills. This research posits that our brain
processes motor learning separately from other conscious
types of learning — indicating that we may be able to
eventually train users to accept this feedback subconsciously
to learn motor skills. Corrections may become an automatic
muscle reflex, instead of a conscious mediation.
III. SYSTEM IMPLEMENTATION
Our system for motor learning is made up of optical
tracking, tactile actuators, feedback software, and hardware
for output control. These systems are each described below.
For a more thorough explanation, see [13].
A. Vicon Tracking System for Subject Tracking
Vicon Inc. has designed the most accurate motion capture
system currently available commercially, with millimeter
accuracy in 3-d space over a large workspace. It functions
through the use of roughly one dozen high-speed infraredcameras, matched strobes, and custom hardware. Reflectors
placed on subjects are illuminated by the strobes, and the
cameras use filters such that only those frequency reflections
are visible. By calibrating the software to know the cameras’
locations, 3-d models of all markers can be inferred from the
inpidual 2-d camera views, and the overconstraint of many
cameras increases the positional accuracy. Skeletal models of
subjects are made in software with known marker locations,
and those models are given a least-error kinematic fit in
real-time. Joint angle information must then be calculated
from the known joint positions. Figure 2 shows a subject
wearing our motor learning suit with reflecting markers,
and the associated model with known marker positions and
skeletal kinematic fit. The markers’ placement and the cal-
culated joint angles are used to find the five observed joints:
wrist flexion/extension, wrist abduction/adduction, forearm
rotation, elbow flexion/extension, and upper arm rotation.
Because all subjects have slightly different joint lengths
and offsets, but the same skeletal structure, the Vicon soft-
ware provides a calibration routine to be run on each subject.
The subject runs through a set of movements as they are
tracked, and the software finds the joint lengths and marker
offsets to match the subject to the skeletal template in
order to minimize error. Figure 3 shows the results of the
calibration on one subject, comparing the template nearest
fit to the actual marker positions.
It is worth noting that the Vicon system is very expensive
and bulky. It is not intended as a final solution to the body
tracking problem, as it is too expensive to be practical for
most users. However, it offers the accuracy needed for a
proof of concept, and was an available tool for our initial
experiments. In the future, it is likely that centralized sensors
such as gyros and accelerometers would provide such bodily
measurements without such a prohibitive expense.
B. Tactaid Tactile Actuators
Vibrotactile actuation was chosen as a feedback mecha-
nism for several reasons. Torque for feedback on joints iscumbersome and requires higher power, lowering portability
for use in real learning environments (e.g., a dance class). 动觉触觉交互学习系统英文文献和中文翻译(3):http://www.751com.cn/fanyi/lunwen_15697.html