A robot alone is already capable of performing several complex functions. Now, imagine machines working together, cooperating with each other in a self-learning system that adapts to any situation. This idea was put into practice by researchers from the AMOLF Institute, in the Netherlands.
The scientists set up a group with small autonomous robots, built in the simplest possible way, and connected these little robots in line to see what the experience acquired individually by them could bring to the benefit of the group. The result was surprising.
They were able to obtain interconnected robotic carts that move on trails quickly and steadily. The robots do this without being programmed with a specific route and without knowing what the other robots are doing.
“This is a new way of thinking about the design of self-learning robots. Unlike most traditional robots, this type of simple robot does not require any complex model to allow it to adapt to an environment with any type of change,” he explains. Professor Bas Overvelde.
The secret of union
The self-learning system has several building blocks connecting the robots individually. Each robot has a microcomputer, a motion sensor, an air pump with a bellows and a needle to release that air.
When connecting the robots through the bellows, the displacement of air causes the robotic train to move. The only thing the researchers do in advance is to program a simple command line in each robot, such as turning the air pump on and off every few seconds.
An internal chip in the robots measures speed continuously, allowing them to make small adjustments when turning on the air pump to determine whether the train will move more quickly in a given direction.
“If you allow two or more robots to push and pull each other in this way, the train will move in one direction, sooner or later. Consequently, robots learn that this is the best configuration for their pump without the need to communicate and without precise programming of how to move forward ”, says Professor Overvelde.
The researchers used an algorithm that measures the last speed recorded on each robot to determine the best time for the air pump to start. This system makes it possible for robots to handle situations without having to program them in advance.
“When you can find the right algorithm, this simple system becomes very robust and reliable. Invariably, he will be able to deal with a series of unexpected situations whenever he encounters a new obstacle ahead “, says Professor Overvelde.
Similar experiments were carried out by researchers at the Georgia Institute of Technology in the USA. They used “dumb” robots to perform complex tasks working together. Even without sensors, communication devices or computational memory, the machines managed to work collectively using only their physical characteristics.
But now, Dutch scientists have taken it a step further, adding more elements to this adaptability. They purposely damaged one of the robots, tearing off the needle that releases air into the bellows just to see how the entire system would recover.
According to the team, the robots seemed to have come to life by adapting the whole behavior of the set so that the train continued to move in the right direction, giving further proof that the system is robust and can be applied in any situation.
In the first tests, the researchers were able to produce a mobile train with seven connected robots. The idea now is to increase this number and try other configurations, with more complex behaviors than moving back and forth. “We are going to build an octopus-like system to see how the individual blocks will interact, using a decentralized and fully autonomous“ nervous system ”, adds Professor Overvelde.