IEEE Robotics & Automation Magazine - June 2023 - 48

reduces the influence of critical fast adaptation limiting factors.
Examples of these factors are manual gripper-finger design, production,
mounting, and exchange, as well as limited designer
skills and knowledge of traditional assembly line adaptation
workflows. Accordingly, the described and evaluated concepts
could potentially be adopted and further developed for many different
research and industry applications to enable fast, scalable,
and fully customized small-batch production.
In addition, the proposed automatic fingertip design framework
could potentially be used for automatic training-data
generation and input-data processing in the context of machine
learning-based fingertip design approaches. To illustrate this,
we set up a neural network-based training and testing process
for a single manipulation object, which could potentially speed
up the Bézier surface-based design method.
AUTOMATIC FINGER DESIGN, PRODUCTION,
AND APPLICATION TESTBED
We consider the scenario that an industrial assembly cell should
adapt quickly to a new product with unknown components. To
grasp and manipulate the product's components successfully,
the fingertips of the robot's parallel gripper must be adapted to
the object's characteristics, such as geometry or texture [2].
Accordingly, our approach must have a series of modules
and capabilities in the areas of design, production, and toolchanging
to be able to conduct this gripper-finger adaptation
process. The production, tool changing, task execution, and
evaluation modules have been presented in our previous work
[17]. This testbed setup consists of a production and task-execution
unit with overall two robot arms, 3D printers, fingerbase
magazines, three quick finger-exchange magazines, one
Internet of Things (IoT)-box with three different pick-and-plug
scenarios (key, battery, and ethernet cable) [18], and a graspstability
test setup (Figure 2). The combination of our previously
developed automatic production and application testbed
Gripper Finger
Base Set
Production Setups
[17] with the newly implemented fingertip-design module of
this work results in a new pipeline with the following steps:
1) As soon as a new object should be manipulated by the task
execution unit, the fingertip-design module creates a geometrical
representation of the required fingertip body.
2) The robot arm of the production unit then picks finger
bases from the finger base magazines and loads both printers
with those.
3) The production unit automatically slices the geometrical
representation of the desired fingertip (STL file) and
uploads it to both printers.
4) The fused deposition modeling (FDM) printers print the fingertip
on top of the finger base surface. Note that decoupling
the fingertip from the finger base and printing the fingertip on
top of a standardized finger base reduces the printing and production
time and avoids the need for any support structures,
which a worker would need to remove manually afterward.
5) After the printing process is finished, the finger gets
picked by the robot arm of the production unit and inserted
into their designated quick finger-exchange magazines.
6) The robot arm of the task execution unit can then automatically
mount the newly produced fingers using a specially
designed quick-exchange mechanism. This system enables
the robot to mount and demount fingers on the fly as
desired, also during an ongoing assembly task.
7) After mounting the new finger-pair, the robot can conduct
the previously mentioned manipulation tasks using the
IoT-box setup [18] as well as the grasp-stability tests via
the corresponding test-ring infrastructure.
Finger
Design
Pipeline
Manipulation
Request
?
AUTOMATIC FINGERTIP DESIGN PIPELINE
To conduct the desired automatic fingertip design in step one
of the described testbed approach, a design subpipeline is
required. This pipeline contains multiple steps to derive the fingertip
block geometry based on a given object. To grasp these
objects successfully, the fingertips must
often provide a special form, which at
least partially encloses the desired
object [2]. Especially for spherical
objects, friction-closure-based grasping
approaches are often insufficient to
establish a stable grip.
Accordingly, the proposed design
Final Result
Successful
Manipulation
FIGURE 1. Automatic fingertip design, production, and task execution setup. A new pair
of gripper fingers can be automatically designed and produced on-demand based on a
given manipulation scenario and related object. The produced fingers can be automatically
loaded by a designated robot arm and applied to conduct the desired task.
48 IEEE ROBOTICS & AUTOMATION MAGAZINE JUNE 2023
approach aims at maximizing the object
covering. We implemented three different
approaches to realize such a form-closure-based
design method: a projected
surface, a Bézier surface fit optimizationbased,
and a machine learning-based
design approach. All three methods complete
each other and have their individual
advantages and disadvantages, which
make them interesting to compare for
potential integration into our setup. For
example, the projected surface approach
is fast to compute and mimics a direct
Scale
IEEE Robotics & Automation Magazine - June 2023

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2023
