IEEE Robotics & Automation Magazine - December 2017 - 82

Normal Gait

F
E

E
F
F

E

E

F

Dh

50
30
20
10

Flexion

0
-10

(b)

90
80
70
60
50
40
30
20
10
0
-10

Swing

40

Hf
(a)

Stance

Knee-Joint Angle (°)

Hip-Joint Angle (°)

60

WalkON Suit

Extension
0 10 20 30 40 50 60 70 80 90 100
Gait Cycle (%)
(c)

Stance

Swing

Flexion
Extension
0 10 20 30 40 50 60 70 80 90 100
Gait Cycle (%)
(d)

Figure 5. The schematic of the FIW and the joint-angle trajectories for a gait speed of 18 m/min: (a) the height difference between
the leading and trailing leg, (b) a comparison with normal walking, (c) the hip-joint-angle trajectory ( i r,hip ), and (d) the knee-jointangle trajectory ( i r,knee ). E: extension; F: flexion.

Intention
Recognition

Motion xd, zd
Trajectory
Generation

Body Weight and Dynamic Load

Gait Phase

C

I.K

θr,hip
θr,knee
+

-

Motion
Controller

uhip
+
uknee
+

G

θ hip
θ knee

F.K

x, z

Figure 6. A motion control system for each leg. I.K: inverse kinematics; G: the system model of the WalkON Suit; F.K: forward kinematics.

propel the body forward in a stance phase and guarantee a
large clearance in a swing phase.
Motion Control Algorithms
A motion-control algorithm is depicted in Figure 6. Once a
command is executed by the pilot using the switches on the
crutches, the desired positions of the feet, x d and zd, are
determined in real time. Then, the desired joint-angle trajectories, i r,hip and i r,knee, are calculated by the inverse
kinematics of the robotic leg, i.e., I.K in the figure. The
control input for each joint actuator, u hip (k) and u knee (k), is
calculated as
u (k) = f1 i r (k) + f2 i r (k - 1) + c 1 e (k)
+ c 2 e (k - 1) + c 3 e c (k),

(4)

Table 3. The settings of the controller gains.

82

*

Phase

f1, f2

c1, c2

c3

s

Stance

0

170,000, 4,000

0

0

Swing

45,000, 950

6,500, 150

500,000

1

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

DECEMBER 2017

where k is a current discretized time and e (k) = i r (k) - i (k)
is a tracking error for each joint. The subscripts, hip and
knee, are omitted for simplicity. f1 and f2 are feedforward
controller gains, and c 1, c 2, and c 3 are feedback controller
gains. e c (k) is an accumulated error, i.e.,
e c (k) = s [e c (k - 1) + e (k) T],

(5)

where T is the sampling period and s is a switching variable.
The value e c (k) is reset to zero if s = 0.
During the operation of the WalkON Suit, the controller
gains were changed according to the gait phase. In the stance
phase, the robotic legs are exposed to large disturbances due
to the body weight and dynamic load. Also, the disturbances
are continuously changed according to walking conditions,
such as speed, ground condition, and crutch position. Therefore, high gain control is not preferred because the control
input is drastically altered according to the change of the disturbance, which causes an instability of the overall control
system and makes the pilot uncomfortable and nervous.
Therefore, the robotic leg was controlled by the feedback control gains related to the proportional and derivative control,
i.e., c 1 and c 2, while the other control actions depend on the
phases, as described in Table 3.
Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2017
