This study was led by the team from the Bio-inspired Soft Robotics Laboratory at the Italian Institute of Technology (IIT). They drew inspiration from the octopus's nervous system: although an octopus has a relatively small central brain, about 60% of its neurons are spread across its eight arms, with each arm capable of processing information on its own and triggering reflex actions—like catching prey independently without waiting for orders from the brain. The research team tried to replicate this distributed architecture in a robot system using silicone and electronic components, so that sensing and actions are closely integrated into the flexible 'body' itself, rather than relying on a single central processor.  

The result is a soft robotic arm about 41 centimeters long with a base diameter of around 4 centimeters. Its shape and structure resemble an octopus arm, and it’s equipped with 10 artificial 'suckers' from base to tip, gradually decreasing in size. The system is designed without relying on cameras, external computers, or a centralized control unit, instead putting core sensing and basic decision-making capabilities directly inside the suckers. This research has been published in the journal *Nature Machine Intelligence*, and IIT has also released public-facing introduction materials.  

Each artificial sucker integrates three LEDs and three phototransistors, forming a mini optical sensing system that detects changes in reflected light, essentially acting as a local nerve node for this 'mechanical arm.' When something touches the surface of a sucker, the silicone deforms, changing the light path, allowing the system to determine if contact has occurred, the force of the contact, and the direction of the incoming object—three key pieces of sensory data. Tests show the system’s force sensitivity is about 400 millivolts per newton, with a force measurement error of roughly 0.1 newtons, roughly the weight of a few paperclips; the maximum error in direction recognition is under 18 degrees, and the average error is around 8 degrees, close to the angle between adjacent marks on a clock.  

In terms of control architecture, this soft robotic arm uses a two-level system: the first level is fully local—each sucker has its own circuit and triggers adhesion immediately upon detecting contact, without waiting for central commands; the second level is at a higher layer, collecting data from all suckers and analyzing the target position and contact features over about a 4-second window, deciding the overall grasping strategy, like bending the arm up or down, rotating it, and overriding local sucker actions when needed. The team noted that integrating sensing and signal processing directly into each sucker allows the arm to react in real time and with precision without centralized control, featuring good scalability and robustness, capable of operating in complex environments, including underwater.

All the experiments are currently being conducted underwater. During testing, this robotic arm can detect objects like glass bottles and glasses while moving, estimate the weight of the object at about 72.5 grams (the actual weight is 85 grams), and handle targets at various angles, including an artificial 'starfish.' In terms of load capacity, the arm can lift objects weighing up to around 500 grams. Its sensing performance remains stable even after 300 cycles of repeated use, showing good durability. Because each suction cup only sends refined information like the direction of contact to the upper control unit, rather than all the raw data, the system's bandwidth requirements are greatly reduced, making it easy to scale up with more suction cups or even multiple tentacles without noticeably sacrificing response speed.  

The research team pointed out that the design is highly modular, and the number and arrangement of suction cups can be flexibly adjusted depending on the task. Potential applications include inspecting underwater infrastructure such as pipelines, cables, and platforms, as well as collecting biological samples in narrow or complex areas that rigid robots can't reach. With its combination of flexible structure and autonomous decision-making capability, this 'octopus-style' robotic arm could offer new technological solutions for deep-sea exploration, marine engineering, and underwater maintenance.  

Octopuses have long been an important source of inspiration for biomimetic design in robotics. As early as 2017, German automation company Festo showcased the OctopusGripper at the Hannover Messe, a silicone tentacle-style gripper powered by compressed air, which wraps around objects via two rows of suction cups when inflated to complete gripping, though it still heavily relied on external air pressure control and manual operation. In recent years, researchers at the University of Bristol in the UK approached the problem differently, focusing less on mimicking the tentacle shape and more on the mucus secreted by octopus suction cups. They developed a new type of suction cup made of multilayer soft structures and bionic fluid systems that can simulate how octopus mucus blocks gaps on rough surfaces, allowing it to firmly grip irregular objects like stones and wood that traditional suction cups struggle with.  

Going even further, research teams from Peking University, the National University of Singapore, Zhejiang University, and Beijing Institute of Technology jointly designed the OUT-Robot mechanical gripping system, which mimics the grasping strategies of cephalopods, allowing it to quickly switch between soft and rigid states to sort and grab objects of different shapes, flexibility, and weights. Compared with these earlier attempts, IIT's new design stands out for its autonomy—it not only grips objects but can also decide on its own how to grip them. Researchers also noted that the objects used in experiments so far have relatively simple geometric shapes, and their next steps will include testing with more complex and varied shapes and weights, as well as incorporating brain-like neuromorphic computing to make the system's information processing closer to the neural circuits of real octopuses.