Imagine a robot that can not only sense pressure, temperature and painful contact, but can also process that information directly where it is detected, learn from it and react quickly—even if part of its artificial skin is damaged.
That is the idea behind a new neuromorphic robotic-skin system developed by Rohit Abraham John and his team. The researchers have created a decentralized artificial nervous system that moves some of the intelligence away from a central processor and directly into the robot's sensing nodes.
The technology combines memtransistors, artificial nociceptors, learning synapses, neurons and self-healing materials. Together, these components allow robotic skin to identify potentially harmful stimuli, associate different types of touch and help trigger an appropriate response.
Why Today's Robot Skin Has a Major Problem
Most robots are designed to operate in controlled environments. They follow programmed instructions and use sensors to understand what is happening around them.
But future robots are expected to work in much more unpredictable environments. They may need to interact with people, move around obstacles and physically touch objects. For this, robots need something similar to human tactile and somatosensory perception.
Human skin contains enormous numbers of sensory receptors distributed across the body. These receptors detect touch, pressure, temperature and pain. Importantly, the nervous system does not need to send every tiny piece of raw sensory information to the brain before making an initial response.
Robots, however, generally use a different approach.
Sensors placed across the robot collect information and send it through wires to a central processing unit. The processor then analyzes the information and decides what the robot should do.
When thousands or millions of sensing points are involved, this creates serious challenges.
Large amounts of data must travel through the system, increasing wiring requirements, communication bandwidth, processing time and energy consumption. Damage to wires or sensors can also affect the system's reliability.
Moving Intelligence to the Sensors
Rohit Abraham John and his team propose a different strategy: instead of sending all sensory information to a central processor, the sensing nodes themselves should perform some of the processing.
This is known as decentralized neuromorphic processing.
The idea is inspired by biological nervous systems, where sensory information can be processed at different levels before reaching the brain.
The researchers designed a three-stage decision-making system.
First, artificial nociceptors identify potentially harmful stimuli and filter information based on short-term changes.
Second, artificial synapses learn relationships between different sensory signals.
Finally, artificial neurons combine this information and contribute to a decision.
This means that the robot does not have to continuously send every low-level sensory signal to a central computer.
A Robot Skin That Can Detect Pain
One of the most interesting aspects of the research is the development of an artificial nociceptor, called a STAR.
In biology, nociceptors are sensory receptors that respond to potentially damaging stimuli. They are associated with the sensation of pain.
The artificial STAR devices use three-terminal indium–tungsten oxide (IWO) memtransistors with ionic dielectrics.
These devices can reproduce a behavior similar to the changing threshold of biological nociceptors.
Instead of reacting equally to every small signal, the artificial nociceptor can filter less important information and respond more strongly when a stimulus crosses an appropriate threshold.
This filtering happens directly at the sensing location.
That is important because it reduces the amount of unnecessary information that needs to travel toward the central processor.
The System Can Learn Associations
Detecting pain alone is not enough for an intelligent robot.
Humans naturally combine different sensory signals. For example, touching an object may produce pressure information, while touching a sharp object can simultaneously produce a painful signal.
The researchers therefore incorporated artificial synapses capable of associative learning.
These devices, called SWARMs, can learn patterns in sensory signals.
In the demonstration, pressure and pain information could be associated with one another. This creates a more intelligent response than simply detecting individual sensor readings.
The learning process is performed close to the sensing nodes rather than being completely dependent on a centralized processor.
This approach can reduce data transfer and potentially make large-scale robotic skins easier to build.
What Happens If the Skin Gets Damaged?
Another major challenge for robotic skin is physical damage.
A robot operating around people or in unpredictable environments could easily experience scratches, cuts or other mechanical damage. If one sensor stops working, a conventional system may lose important information.
The researchers addressed this problem using self-healing materials.
Their work demonstrates self-healing neuromorphic devices in which both the mechanical structure and electrical functionality can recover after damage.
Even more interestingly, the system's learning capabilities can help maintain useful signal processing when a nociceptor is damaged after learning.
This gives the artificial nervous system a degree of fault tolerance.
The researchers report that the devices can repeatedly heal at room temperature, including damage occurring at the same location.
That could become particularly valuable for flexible electronics, robotic skin and future prosthetic systems.
Why Memtransistors Matter
Traditional CMOS-based neuromorphic circuits can require many individual electronic components to reproduce the functions of neurons and synapses.
That makes it difficult to create extremely large networks of sensors.
Memristive devices offer another possibility because they can combine memory and computation in compact hardware.
The researchers use three-terminal memtransistors that can be configured to perform different functions.
One configuration acts as an artificial nociceptor, while another can behave like an artificial synapse.
This reduces the number of components required and can simplify wiring.
The devices can also support flexible and potentially stretchable electronics because several of their layers can be fabricated as thin films.
That makes the technology particularly interesting for soft robotics, where conventional rigid electronics can be difficult to integrate.
Testing the Artificial Nervous System
To demonstrate the concept, the team developed a proof-of-concept system containing artificial nociceptors along with receptors capable of detecting pressure and temperature.
These sensing elements were connected to memristive learning synapses and CMOS neurons.
The system was able to distinguish harmful contact, such as contact with a sharp tip, while also processing pressure and temperature information.
The researchers used these signals to demonstrate an association between pressure and pain perception and to produce an escape-like response in a sensorized robotic arm.
The important point is not simply that the robot can detect pain-like signals. It is that some of the decision-making happens directly at the sensor level.
A Step Toward an Artificial Nervous System
The researchers' approach represents a significant shift from the traditional sensor-to-central-processor model.
Instead of treating robotic skin as a huge collection of sensors that continuously report everything to a central computer, the skin itself can become part of the robot's computing system.
This could reduce data transfer, latency, wiring complexity and energy requirements while improving robustness.
The concept also has similarities to biological systems. Some rapid protective responses in animals can be processed at lower levels of the nervous system rather than waiting for complete conscious processing in the brain.
Of course, the technology is still at a prototype stage. It does not mean robots currently experience pain in the same way humans do. The artificial nociceptors reproduce specific physical and computational behaviors associated with biological pain sensing.
Nevertheless, the research provides an important framework for building intelligent, flexible and self-repairing robotic skin.
In the future, similar technology could be extended beyond pain and touch to other sensory systems. Robots could potentially have distributed networks that detect environmental changes, learn locally and respond rapidly without constantly depending on a powerful central computer.
The long-term vision is remarkable: a robot whose skin does more than sense the world—it processes information, learns, responds and potentially repairs itself.
That could bring robotics one step closer to machines with an artificial peripheral nervous system, making future robots and prosthetic devices more adaptive, resilient and capable of safely interacting with the real world.
Reference: John, R.A., Tiwari, N., Patdillah, M.I.B. et al. Self healable neuromorphic memtransistor elements for decentralized sensory signal processing in robotics. Nat Commun 11, 4030 (2020). https://doi.org/10.1038/s41467-020-17870-6
Comments
Post a Comment