To
reliably complete household chores, assemble products and tackle other manual
tasks, robots should be able to adapt their manipulation strategies based on
the objects they are working with, similarly to how humans leverage information
they gain via the sense of touch. While humans attain tactile information via
nerves in their skin and muscles, robots rely on sensors, devices that sense
their surroundings and pick up specific physical signals.
Most robotic hands and grippers
developed so far rely on visual-tactile sensors, systems that use small cameras
to capture images, while also picking up surface deformations resulting from
contact with specific objects.
A key limitation of these sensors is
that they need to be made of stiff materials, to ensure that the cameras
capture high-quality images. This reduces the overall flexibility of robots
that rely on the sensors, making it harder for them to handle fragile and
unevenly shaped objects.
Researchers at Zhejiang University
recently introduced FlexiRay, a new robotic hand that could overcome this
limitation via the combination of an innovative mechanical structure and deep
learning algorithms.
The robotic hand, presented in a paper published in Nature
Communications, is based on a flexible structure that naturally bends
inward when it is squeezed, producing deformation patterns that are then
analyzed by a deep learning algorithm.
"The inspiration for this work came
from the remarkable capabilities of the human hand, which combines soft,
compliant skin with a complex sensory system capable of perceiving pressure,
texture, and temperature simultaneously," Huixu Dong, senior author of the
paper, told Tech Xplore.
"While soft robotic systems have made great strides in structural compliance (safety and adaptability), integrating high-resolution sensing into these soft bodies has been a major challenge. Our primary objective was to solve the 'blind spot' problem in soft sensors."
Human-Robot Interactions. Credit: Nature
Communications (2025). DOI: 10.1038/s41467-025-67148-y
FlexiRay: A human-inspired robotic gripper
When reviewing earlier robotics
literature, Dong and his colleagues realized that most available visual-tactile
sensors are not very flexible, as substantial surface deformations can
adversely impact the field of view of embedded cameras. They thus set out to
design an alternative gripper that can sense objects with high precision, even
when it is significantly deformed, twisted or bent around objects.
"FlexiRay is a flexible
robotic finger that 'sees' what it touches," explained Dong. "It is
built on a bio-inspired structure called the Fin Ray Effect, which allows it to
passively wrap around objects like a fish fin. It works using an internal
camera and a unique 'multi-mirror' optical system."
Typically, when a soft robotic
finger bends, it blocks the field of view of cameras embedded inside it. The
team's gripper, on the other hand, has a unique internal optical architecture
that allows it to dynamically redirect the camera's field of view in response
to the passive deformation of its fingers.
"Essentially, the mechanical
deformation of the finger drives the optical system to 'look around' the
corners," said Dong. "This allows FlexiRay to wrap around irregular
objects (like a fish fin) while maintaining a continuous, unobstructed view of
the contact surface.
"It transforms the structural
obstruction from a hindrance into a functional mechanism for full-coverage
sensing. Its 'skin' is a multi-layered pad containing thermochromic
(color-changing with heat) and reflective materials."
The primary
advantage of FlexiRay is that it can reliably collect visual and tactile
information even if it is built using very flexible materials. With a single
camera, the robotic gripper can precisely detect an applied force, its own
shape and where it is touching an object, as well as the object's texture and
temperature.
"In initial tests, our robotic hand achieved over 90% effective sensing coverage even during large deformations," said Dong. "The most notable achievement of our work is the paradigm shift from 'avoiding deformation' to 'leveraging deformation.' We have shown that a robotic finger can achieve the large passive deformations required for safe, compliant grasping without sacrificing sensory perception."
Allowing robots to manipulate objects more reliably
The newly introduced robotic hand
has so far achieved very promising results, which highlight its potential for
the manipulation of delicate objects and tools that robots often handle poorly.
The team's prototype gripper could soon be improved further or integrated with
other components to realize systems tailored for specific real-world
applications.
"Unlike rigid sensors that
perceive only a small, flat patch, FlexiRay achieves a massive effective
sensing area because it conforms to the object's shape," said Dong.
"By converting large
structural deformations into valid optical signals, we successfully replicated
five human-like senses (force, location, texture, temperature, and
proprioception) in a single, robust soft body."
In the future, FlexiRay could prove
advantageous for the manipulation of various real-world objects, ranging from
delicate agricultural products to irregular packages. As it is based on softer
materials than many robotic hands introduced in the past, it could be safer
when deployed in proximity to humans or objects that can be easily damaged.
"Currently, FlexiRay operates
as a gripper, but our future research aims to expand this technology into fully
multi-fingered robotic hands to enable even more complex manipulation,"
added Dong.
"We also plan to integrate this hardware with imitation learning frameworks. This would allow robots to learn dexterity and safety directly from human demonstrations, using the rich tactile and thermal data FlexiRay provides to better understand the physical world."
Source: Soft robotic hand 'sees' around corners to achieve human-like touch
No comments:
Post a Comment