IEEE Robotics & Automation Magazine - March 2013 - 59

Case Study 2: Amputee Control of a Robot Arm
Having demonstrated the applicability of GVFs for predicting
and anticipating sensorimotor signals during able-bodied
subject trials, we next assessed the ability of this approach to
predict robot grip force, actuator position, actuator velocity,
and other sensorimotor signals related to an amputee's interactions with a robotic training prosthesis (Figure 6).
Two trials were conducted with an amputee participant.
The subject was a 20-year-old male with a left transhumeral
amputation, injured in a work accident 16 months prior to the
first trial. Six months prior to the first trial, the subject underwent surgical revision of his limb, involving targeted muscle
reinnervation as described by Dumanian et al. [11], as well as
targeted sensory reinnervation. The motor reinnervation procedure involved rerouting of the median nerve to innervate
the medial biceps remnant muscle and the distal radial nerve
to the lateral triceps muscle to provide additional myoelectric
signal control sites for myoelectric prosthesis control. Sensory
reinneration involved identifying specific median nerve fascicles with high sensory content, and coapting these fascicles to
the intercostal brachial cutaneous sensory nerve. In a similar

NMARE (V)

NMARE (Radians)

trial runs by one of the subjects. Learning
0.8
0.30
progress for joint angle prediction and myo0.7
electric signal prediction is shown in terms of
0.25
0.6
the NMARE for the 0.8 s time scale, averaged
0.20
EMG
0.5
into 20 bins; error bars indicate the standard
0.4
0.15
Elbow
0.3
deviation ( v ) over eight independent trials.
0.10
0.2
After 5 min of learning, the average NMARE
0.05
0.1
Hand
for both actuators was less than 0.2 rad (11.5°)
0.0
0.00
0
1
2
3
4
5
0
1
2
3
4
5
for the 0.8 s time scale and 0.3 rad (17.2°) for
Time
(min)
Time
(min)
the 2.5 s time scale. For the prediction of myo(a)
(b)
electric signals, the average NMARE was less
than 0.15 V, or 3% of the maximum signal
magnitude (5 V). Prediction performance Figure 5. Learning performance during able-bodied subject trials. Prediction
continued to improve over the course of an learning curves are shown for (a) elbow and hand actuator position and (b) the
average of all four myoelectric signals. Accuracy is reported in terms of NMARE
extended learning episode. These results are averaged over eight trials.
representative of the learning curves plotted
for the other participants.
fashion, ulnar sensory fascicles with high sensory content
We also observed the adaptation of learned predictions in were coapted to the sensory branch of the axillary nerve, and
response to real-time changes. Perturbations to the task envi- the remainder of the ulnar nerve trunk was rerouted to the
ronment led to an immediate decrease in prediction accuracy, motor branch of the brachialis muscle. Reinnervation resulted
followed by a gradual recovery period. After a short learning in a widely distributed discrete representation of digital sensaperiod with one able-bodied subject, we transferred the four tion on the upper arm. A second trial occurred nine months
EMG electrodes to comparable locations on a second able- after the first, wherein we observed noticeable changes to the
bodied subject. Within a period of less than 5 min of use by subject's nerve reinnervation and improvements in his ability
the new subject, the system had adapted its predictions about to operate a myoelectric device.
joint motion to reflect the new individual. Similar results were
The robot platform used by this subject was the myoelecfound for tasks involving gradual or sudden muscle fatigue; tric training tool (MTT), a clinical system designed to help
prediction accuracy was found to remain stable during peri- new amputees prepare for powered prosthesis use [6]. The
ods of extended use (>60 min of activity). Predictions were MTT includes a five-degree-of-freedom robot arm that mimalso found to recover their accuracy when a subject began ics the functionality of commercial myoelectric prostheses.
holding a moderate weight with their controlling arm part of
the way through a session. These observations reflect preliminary results, and further study is needed to verify the rate and
Force Sensor
amount of adaptation achievable in these situations.
Robotic
Gripper

(a)

(b)

EMG
Electrodes Tactor

Force Sensor
Ball

Gripper

MTT
EMG
Controller

Top View
Side View
Gripper

(c)
Figure 6. The experimental setup for amputee trials: (a) a multijoint
robot arm and force sensor, (b) a participant with Bagnoli-8 EMG
system, and (c) a schematic of the MTT and tactor feedback system.

March 2013

IEEE ROBOTICS & AUTOMATION MAGAZINE

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - March 2013