Mazes are commonly used in psychology to assess the behavior of rats and mice. As scientists create more and more human-like robot brains, they thought it was time for the machines to have a turn.
Robot vs. maze
Teams from the Eindhoven University of Technology and the Max Planck Institute for Polymer Research, paired up to find the interesting answer to this query. They designed an algorithm for the robot so it made directional decisions like it had a human brain, utilizing machine learning neural networks. Every time the robot was to make a correct turn, a certain amount of electricity was put through the machine. This idea is modeled from biological synapses, which are strengthened each time information transmits through them.
The AI robot is made from the LEGO Mindstorms EV3 robot kit. The team equipped it with two wheels, traditional guidance software, plus touch and reflectance sensors. It was then deployed in a two meter by two meter honeycomb shaped maze. The robot was let loose in the structure, navigating around until it was able to escape.
Who came out on top?
“In the end, it took our robot 16 runs to find the exit successfully,” says Imke Krauhausen, a Ph.D. student working on the project. “And, what’s more, once it has learned to navigate this specific route, it can navigate any other path that it is given in one go. So the knowledge it has acquired is generalizable.”
A key feature that drove the robot’s success was its combination of programmed sense and movement. This idea was modeled again from nature where they strengthened memory and learning, reinforcing one another.
Why is this important?
This new study, published in Science Advances, paves the way for exciting new applications of neural networks. These include areas in medicine, energy conservation, data storage, e-commerce, security, and loan applications, just to name a few.
A growing issue in this field is energy output. These neural systems take a large amount of energy to be trained and operate, new innovative ideas are being invented to combat this such as modeling processing systems from astrocytes, a type of star-shaped brain cell. When this issue is overcome, the power and application of these networks will be incredible.
Another groundbreaking front of this research was the organic polymers the scientists constructed the robot out of, in order for the joint sense-memory feature to work. It has huge applications in the future of neural networking, plus in biomedical applications. Surprisingly this material may be able to be integrated with actual nerve cells, allowing amputees to regain feeling in bionic hands.
Source study: Science Advances – Organic neuromorphic electronics for sensorimotor integration and learning in robotics