A human-like brain helps a robot to get out of a maze autonomously: the innovation confirms that organic neuromorphic robots can not only learn, but are also capable of moving independently and solving a complex problem.
A group of researchers from the Eindhoven University of Technology in the Netherlands and the Max Planck Institute for Polymer Research in Mainz, Germany, created a robot that bases its decisions on an organic brain with a functioning structure similar to the human brain. The study paves the way for exciting new applications of neuromorphic devices in healthcare and other fields.
Advances made in recent years in devices based on machine learning, artificial neural networks and other Artificial Intelligence applications are undoubted. However, these approaches come up against a major disadvantage when they want to match the efficiency of the human brain: algorithms require enormous amounts of energy to carry out their processes, whereas our brain works faster and better with a minimum energy flow.
Mimicking the human brain
The energy efficiency of the human brain originates in the way neurons work: they communicate with each other through so-called synapses, which get stronger every time information flows. This “plasticity” and capacity for permanent modification is what ensures processes such as learning and memory, among others.
In the late 1980s, Carver Mead developed the concept of neuromorphic engineering. This is the use of a large-scale integration system contained in analog circuits, with the purpose of mimicking neurobiological structures located in the human nervous system. In other words, the idea is to develop technological and organic devices that “copy” the functional structure of the human brain, thus gaining in efficiency and getting closer to our worldview.
More than 30 years later, neuromorphic robots are already a reality: it is even known that they have the ability to learn just like human beings, but the latest innovations are also demonstrating that they can mobilize themselves independently in order to achieve a goal. According to a press release, the engineers and scientists in charge of the new study, published in the journal Science Advances, have confirmed the development of a neuromorphic robot that manages to get out of a maze on its own.
Memory, senses and movement
It is common for mazes to be used to evaluate learning abilities, for example in the case of experiments with rodents. In view of this, the specialists wanted to test whether a neuromorphic robot could learn in the same way the turns and movements necessary to escape from the maze. This was achieved with autonomous movements, making it clear that the potential of these devices seems to have no limits.
The researchers used a robot called Lego Mindstorms EV3, equipped with an organic neuromorphic brain. The device was programmed to turn right by default, but when it encountered a dead end or deviated from the designated path, a signal told it to turn back or turn left. Subsequently, a neuromorphic device integrated into the robot recorded the corrective stimulus, so that the machine could remember it in its next movements.
After 16 attempts, the robot managed to “memorize” all possible exits and was able to follow a successful route independently. By demonstrating that the artifact can efficiently navigate any other route given to it, the researchers proved that the knowledge acquired by the robot is generalizable.
Furthermore, the experts remarked that the robot’s ability to learn and get out of the maze lies mainly in the unique integration between sensors and motors: such an alliance, in which sense and motion reinforce each other, is also the way nature operates. The applications for these new devices span multiple fields, although their use in the area of artificial prostheses, organ replacement and other sectors of regenerative medicine, for example, stands out.