'Human-like' intelligence helps robots escape the maze

Machine learning and neural networks have been extremely popular in recent years, which is understandable given their numerous achievements in image identification, medical diagnosis, e-commerce, and a variety of other sectors. However, this software-based approach to machine intelligence has downsides, not the least of which is that it uses so much power.

Simulating the human brain

This power issue is one of the reasons why academics have been working to design computers that are significantly more energy efficient. And to discover a solution, many people are looking to the human brain, a thinking machine that is unrivaled in its low power consumption due to the way it mixes memory and computation.

Neurons in our brain connect with one another via synapses, which are strengthened each time information goes through them. This flexibility ensures that people remember and learn.

"In our research, we used this approach to create a robot that can learn to travel around a labyrinth," explains Imke Krauhausen, PhD student at TU/Mechanical e's Engineering department and lead author of the paper.

"A synapse in a mouse brain is strengthened each time it takes the correct turn in a psychologist's maze, and our gadget is 'tuned' by providing a specific quantity of electricity. You can modify the voltage that controls the motors by adjusting the resistance in the device. They, in turn, determine whether the robot turns left or right."|

So, how exactly does it work?

Krauhausen and her colleagues conducted their investigation with a Mindstorms EV3, a Lego robotics kit. It was placed into a 2 m2 huge maze made up of black-lined hexagons in a honeycomb-like layout, equipped with two wheels, typical guiding software to ensure it can follow a line, and a variety of reflectance and touch sensors.

By default, the robot is configured to turn right. When it comes to a dead end or deviates from the designated path to the exit (as indicated by visual signals), it is instructed to either return or turn left. This corrected stimulus is subsequently stored in the neuromorphic device for future use.

"In the end, it took our robot 16 runs to correctly identify the exit," Krauhausen explains. "Furthermore, once it has trained to navigate this specific route (target path 1), it can navigate any other road that is presented to it in a single pass (target path 2). As a result, the knowledge it has gained is generalizable."

According to Krauhausen, who collaborated closely with the Max Planck Institute for Polymer Research in Mainz on this research, the unique integration of sensors and motors contributes to the robot's capacity to learn and exit the maze. "This sensorimotor integration, in which sense and movement reinforce one another, is also very much how nature operates, so we attempted to recreate it in our robot."

Smart Polymers

The organic material employed for the neuromorphic robot is another brilliant aspect of the research. This polymer (known as p(g2T-TT)) is not only stable, but it can also'retain' a considerable portion of the specific states that it has been tuned to during the multiple runs through the labyrinth. This ensures that the taught behavior'sticks,' similar to how neurons and synapses in the human brain recall events or actions.

Paschalis Gkoupidenis of the Max Planck Institute for Polymer Research in Mainz and Yoeri van de Burgt of TU/e, both co-authors of the research, pioneered the use of polymer instead of silicon in the field of neuromorphic computing.

In their research (from 2015 to 2017), they demonstrated that the material can be tuned in a considerably wider range of conduction than inorganic materials, and that it can'remember' or preserve learned states for extended periods of time. Since then, organic devices have been a hot topic in the field of hardware-based artificial neural networks.

Bionic hands

Polymeric materials also offer the added benefit of being used in a wide range of biomedical applications. "Because of their organic origin, these smart gadgets can theoretically be integrated with genuine nerve cells." Assume you lost your arm as a result of an injury. Then you might perhaps utilize these gadgets to connect your body to a bionic hand," says Krauhausen.

Another promising application of organic neuromorphic computing is in small, so-called edge computing devices that process data from sensors locally, rather than in the cloud. "This is where I see our devices going in the future," says Van de Burgt, "our materials will be very useful because they are easy to tune, use much less power, and are inexpensive to make."

So, will neuromorphic robots be able to play soccer like TU/soccer e's robots one day?

Krauhausen (Germany): "In theory, that is certainly possible. But there is still a long way to go. To move around, our robots still rely on traditional software. And, in order for neuromorphic robots to perform extremely complex tasks, we must construct neuromorphic networks in which many devices collaborate in a grid. That is something I plan to work on during the next phase of my PhD research."

Watch the video : https://youtu.be/O05YVljxrtg

Post a Comment

Previous Post Next Post