IEEE Robotics & Automation Magazine - June 2021 - 108

implemented graphically without having to modify an existing
method.
For instance, we upgraded a plain pick-and-place system
so that placement into a crate is attempted only if the picking
stage succeeded. After integrating a grasping technique
[19], we integrated a high-level method that performs an
image difference to check if an object has been grasped and
a new state that generates a new end-effector pose within a
given radius.
These different scenarios show the variety of tasks and
robots that can be targeted using GRIP by combining different
integration modalities. It is important to note that some of
these tasks could not have been implemented using commercial
solutions (e.g., drag&bot) because the set of supported
robots is fixed.
On the other hand,
The GRIP framework can
also be used to easily
generate new and reactive
behaviors from existing
components at the
task level.
GRIP allows users to interface
their own robot using
widely available materials.
Designing complex tasks,
such as bimanual operation
through textual programming
(e.g., with SMACH),
would be tedious due to
the substantial number of
states and containers in -
volved. While RAFCON
and FlexBE provide ad -
vanced GUIs to design
similar tasks, integrating
new robots or software
components is less flexible. For example, each component to
be run by such frameworks would need to be rewritten in a
specific language following a very specific format or template.
User Study
GRIP's practical usability has been evaluated in a user study.
Its purpose is to assess the intuitiveness of 1) interfacing different
robot hardware, 2) integrating different software components,
and 3) designing and executing manipulation tasks
in typical robotic scenarios.
Experimental Protocol
To evaluate GRIP in realistic conditions, we asked participants
to perform a set of four tasks, in the following sequence:
1) Integrate a UR5 robot arm using MoveIt!, and make it
move between two prerecorded joint states.
2) Change the kinematics library and motion planner configured
by default to TRAC-IK and BiTRRT. Repeat the previous
motion to make sure the integration was successful.
3) Using a launch file, run the Shadow Dexterous Hand
mounted on a UR10 arm, and add the necessary states to
the previous task to create a pick-and-place scenario from
prerecorded poses.
4) Add a Kinect2, and integrate an external grasping algorithm
[19] to run an autonomous pick-and-place task.
108 * IEEE ROBOTICS & AUTOMATION MAGAZINE * JUNE 2021
To avoid unexpected behaviors, participants were asked to
select the joint states and poses from among a list of safe ones.
They were provided with the links to the official repositories
of the grasping algorithm and Shadow Hand. For convenience,
the names of the topics generated by Kinect2 were
also provided in the instructions.
To reflect realistic conditions, participants did not have any
training time with our software but were provided with documentation.
The latter contained only generic principles
(e.g., wrapping code inside an ROS server) and examples for
robots and software components that were not involved in
the experiment (e.g., Franka Panda robot), limiting the documentation
bias.
The cohort was composed of 25 participants: 15 men and
10 women between 22 and 54 years old. All participants had a
computer science background and were first-time users of
GRIP. Among them, only three had experience in programming
Universal Robot arms, and none had worked with the
Shadow Dexterous Hand.
Before the study, each participant was asked to complete a
questionnaire about his or her background. No personal
information was asked (apart from age and gender), and participants
gave their consent that the collected data were going
to be anonymously used in a scientific study. This questionnaire
consists of three 0-4 rating scales corresponding to the
following questions:
●
How familiar are you with robotics? Was it part of your
education, or do you have other experience in the field?
●
●
How familiar are you with programming robot hardware
for grasping and manipulation tasks?
How familiar are you with ROS?
For these three questions, answering zero means having
no knowledge, whereas four means having in-depth knowledge
and, thus, being proficient in the task, tool, or concept. Based on
the replies from participants, we were able to define three groups:
●
●
group A: no knowledge of robotics (9 participants)
group B: basic robotics knowledge but never used ROS
(8 participants)
●
group C: previously used ROS or have previous experience
in robot programming for manipulation tasks (8 participants).
This
classification allows us to estimate the value of GRIP
for each type of user. A useful metric for this is the time
spent by participants on each task and subtask (i.e., configuring
hardware and software components as well as
the design and execution of the task). We also measured
how much time participants spent consulting documentation
and external resources rather than using the
framework itself.
To allow for collection of these data, participants worked
with a version of GRIP that embeds several timers. Every
time a user changed the window focus, time stamps were
stored. In addition, participants were asked to press a button
when they finished a task. The resulting time stamps allowed
us to determine the time spent on windows within and outside
our GUI.
IEEE Robotics & Automation Magazine - June 2021

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2021
