IEEE Robotics & Automation Magazine - December 2018 - 29

Position Error

0.08
0.06
0.04

(m)

0.02
0
-0.02
-0.04
x
y
z

-0.06

0

-0.76
-0.78

0

20
Time (s)

Joint 2 Position

1.15

0

-0.7
-0.8

0

20
Time (s)
(b)

70

Joint 3 Position

-0.5
-1
-1.5

20
Time (s)

-0.6

60

0

Joint 5 Position

-0.5

(rad)

-0.74

50

1.2

1.1

20
Time (s)
Joint 4 Position

-0.72

30
40
Time (s)
(a)

1.25

(rad)

-0.5

-0.6

20

(rad)

10

Joint 1 Position

-0.4

(rad)

0

0

20 40 60
Time (s)
Joint 6 Position

0.1

(rad)

-0.08

(rad)

In the shared control mode (Fig-
ure 5), Gaussian-mixture regression is
used on both the teleoperation and
robot sides to generate probabilistic
trajectory distributions, represented as
a tube in the figure. On the teleopera-
tor side, this tube is adapted locally
to match the situation of the virtual
environment in which the user is
immersed-here, the model can be
used, e.g., for haptic corrections. On
the robot side, the same model adapts
to the situation that is locally detected.
This situation can potentially differ
from the one currently experienced by
the user, as depicted in the figure: the
two tubes have different shapes but
share the same GMM parameters.
In this way, the robot is provided
with a fast adaptation technique that
can directly exploit the locally sensed
information, i.e., the tube is adapted
online without passing through the
slow satellite communication. This
type of assistance is relevant for han-
dling small transmission delays, i.e., to
cope with situational discrepancy due
to the slow refreshing rate. For longer
delays, a semiautonomous mode, as
described next, is usually preferred.
In the semiautonomous control
mode (Figure 6), a linear quadratic
tracking controller in operational
space is used to generate a reference
trajectory starting from the current
robot pose. These acceleration com-
mands in operational space are used
by the robot controller until the teleop-
erator decides to abort the task or
switch to another task. On the teleop-
erator side, the retrieved trajectory is
used for visualization purpose.

0

-0.1

0

20
Time (s)

Figure 7. (a)The position error over time during the experiments. The error is relatively
high because the control gains were maintained at low levels for safety reasons during
these tests. (b) The joint positions and the minimum/maximum thresholds are shown
(in red). The plots show how the TPIK approach enforces the validity of the joint limits.

ROV Control for Intervention Missions with
Communication Latencies
Task Priority Inverse Kinematics
From a control point of view, the DexROV system is much
more similar to an AUV than an ordinary ROV in the sense
that it needs to take care of many control objectives on its
own, with high-level inputs coming from the user through
the CE.
For this reason, following an approach similar to the one
adopted for the Trident [7] and Maris [8] projects, the devel-
oped control is a task priority inverse kinematics (TPIK) algo-
rithm that allows one to set a priority order among several

tasks and find the system velocity vector yo that accomplishes
them simultaneously, at best, following the priority order.
.
.
Given a hierarchy composed of k tasks v 1 fv k, the target
velocity yo can be computed as
.

.

y = y 1 + N 1 yo 2 + g + N 1,k - 1 yo k,

(1)

where each yo i is the velocity contribution of task i, and N 1, i
is the null space of the augmented Jacobian matrices from v 1
to v i . If there are conflicting tasks, the projection of the veloc-
ity contribution of the lower-priority tasks into the null space
of the higher-priority ones guarantees that the priority order
is always respected. In addition, this control framework has
december 2018

*

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

29
IEEE Robotics & Automation Magazine - December 2018

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2018
