“If you use rigid limbs, you have to be super precise and make
sure you have fully defined contact when you pick something
up or when you manipulate it,” explains PhD candidate Robert
Katzschmann, who worked on the project and co-wrote the
paper with Rus, graduate student Bianca Homberg, and postdoc
Mehmet Dogar. The paper and research was presented last fall at
the International Conference on Intelligent Robots and Systems.
The researchers’ tentacle-style gripper design expands to
accommodate an object and grasps radially; however, it has one
additional feature that allows it to accurately pick up objects:
sensors. The robotic hand’s three fingers each have special
sensors that can estimate the size and shape of an object
precisely enough to identify it from a set of several items.
Why Soft Robotics?
The silicone fingers are actually part of a larger area of study
out of Rus’ Distributed Robotics Lab at CSAIL, with the aim
of showcasing the benefits of soft robotics made out of less
conventional materials. For example, in 2014 another graduate
student demonstrated a squishy snake-like robot which can
navigate through tricky mazes.
Due to the soft materials used, these robots can not only
squeeze into tight spaces, but also recover more easily from
collisions and pick up and handle irregularly-shaped objects.
However, because of soft robots’ flexibility, they often struggle
with correctly measuring where an object is, or whether they
actually picked the object up.
That problem is exactly why Rus and her team incorporated
“bend sensors” into the silicone fingers so that they can send
back information on the location and curvature of the object being
grasped. Then, the robot can pick up an unfamiliar object and use
the data to compare to already existing clusters of data points
from past objects.
“By embedding flexible bend sensors into each finger, we got
an idea of how much the finger bends, and we can close the loop
from how much pressure we apply,” says Katzschmann. “In our
case, we were using a piston based closed pneumatic system.”
Currently, the robot can acquire three data points from a single
grasp, meaning the robot’s algorithms can distinguish between
objects which are very similar in size. The researchers hope that
further advances in sensors will someday enable the system to
distinguish between dozens of diverse objects.
“I could potentially see the robots on the factory floor, mostly for
low batch applications, where you can give the factory the ability
for less computation and less adjustment,” remarks Katzschmann.
“It might not be as quick as a specialized gripper, but for smaller
batches you don’t really require that.”
Design & Development
When it’s attached to Rethink Robotics’ Baxter robot, the
gripper far outperforms Baxter’s standard gripper, which couldn’t
pick up a piece of paper or a CD. In fact, the default gripper often
could not handle hollow objects like aluminum cans, crushing
The new robotic hand is further set apart by its ability to
perform both a “pinch grasp,” where the object is held by the
tip of the fingers, and an “enveloping grasp,” where the object is
contained completely within the gripper.
Silicone rubber was chosen for the fingers for several reasons,
but especially for its mechanical characteristics. The material is
relatively stiff, yet flexible enough to bend easily with the pressure
available from the gripper’s actuator pistons. In addition, the
gripper’s interface and exterior finger-molds are 3D-printed,
enabling the system to work on almost any platform.
“The reason why we’ve been using silicone rubber for these
hands is it’s easily available as a two-part solution which you
can mix together, and it cures in a few hours or even minutes
depending upon what mix you use,” explains Katzschmann.
The major characteristic Rus and her team were evaluating in a
material was its elongation-to-break. “You want them to undergo
a lot of cycling with a lot of elongation force in the pleats of the
fingers,” adds Katzschmann. “Each finger pleat has two thin skins
on both sides, and they bubble up a little bit, which is stressful.”
The team had plenty of experience creating molds using
silicone rubber from past projects in the Distributed Robotics Lab.
The real challenge was finding the right sensors that would work
within the system, and then running various experiments to fine-tune some of the parameters.
The Next Generation
Almost immediately after submitting the paper for the gripper’s
first design, the researchers began developing a second-generation design. The paper detailing the new gripper was
submitted for review in January, and the process will take roughly
“With the newer model we’re working on, we have resistance
sensors, but we’re also using force sensors,” says Katzschmann.
“They’re very similar, but instead of just being sensitive to
resistance change, they change their value based on the
compacting of the sensor.”
In addition, the second generation gripper has four digits
instead of three, which allows it to pick up even more objects.
“The challenging part is to pick up small objects that lay flat on
the table, so we have to slide them to the table’s edge and then
pick them up,” explains Katzschmann. “And if you don’t have a
good vision system, and we didn’t initially do anything with vision
for this project, then you’ll have a hard time doing fine positioning
with small objects.”
Rus and her lab will continue to explore the potential of soft
robotics, and how many existing robotic systems could benefit
from adding a few soft elements. For example, when large, rigid
robots fall down, the impact is rather traumatic. Adding a soft
component to the contact point or within the actual joint of the
robot could prove incredibly useful.
“The best part of these soft grippers is that they give you the
flexibility to quickly change the kind of object you would like to
pick up without worrying whether the gripper can deal with it,”
muses Katzschmann. “I think the future of these systems will go
into a direction of combining these fully soft fingers with skeleton