Researchers from AMOLF have landed with autonomous self-learning robots

AI and robotics are going to reign in the coming years. The ongoing digital transformation and technology adoption is the initial stage of this revolutionary step. Robots are already being widely used in many industries for different business operations. These intelligent machines have proved to be a smart addition to the healthcare sector. We witnessed how several companies manufactured robotic disinfectants with the onset of the pandemic. According to a Statista report, the global market for robots is projected to reach USD210 billion by 2025, growing at a CAGR of 26%.

In a recent paper published by researchers at AMOLF's Soft Robotic Matter group, they have shown how self-learning robots can easily adapt to changing circumstances. These small autonomous robots were connected to each other for them to learn on their own to move. They conducted experiments and simulations on a specific robotic platform that could assess and sense their environment and react accordingly. This robotic platform consisted of autonomous units that were identical, as the paper specifies.

Self-learning Robots

Robots that can sense their environment and act according to it are self-learning robots. These robots will not depend on any specific programming as it trains itself to react differently in different scenarios. Autonomously adapting to changing environment and optimizing itself is not an easy task. There have been reports of different methods of achieving self-learning robots. One is intuitive cooperation, which means the robots are connected to AI systems where they can understand speech and act accordingly or intuitively interpret commands from connected devices. It can also be a camera or a sensor that continuously provides images or data. Another method would be shared experiences. Here, robots can be connected to each other and learn from each other's experiences. Robots can be programmed with AI systems and deep learning techniques so that they learn from observing or experiences. XLab had done experiments on developing robots with no codes and instead teach them through shared experience, or by simulating over the cloud. Recently MIT Technology Review covered a pair of robot legs, Cassie that taught himself to walk using reinforcement training.

AMOLF's Self-learning Robots

The research group successfully developed robotic carts that could move on track and learn how to go in a single direction as fast as possible. This did not require any programming or training and it was just through the physical connection of these robots, they learned to move without any centralized information system. The self-learning system comprises small linked building blocks, a microcontroller, a motion sensor, a pump that enables air into the bellows, and a needle to let it out.

They say that this method allows the robots to breathe. By connecting a robot to the other robots below, it will push them to move together. The researchers presented a set of algorithm-based rules to these robots and these tiny machines measured the speed of their action and conducted self-experiments every time the pump is switched on and off.

The group also tried to check the adaptivity of these robots to damage. For this, they purposefully damaged a robot unit by removing the venting needle. But surprisingly, this damage did not stop the robot trail as it started conducting two simultaneous experiments and ket moving ahead. They easily adapted to this sudden change. The speed of these robots is measured by a chip, which is planted inside them. This system is also easy to scale up as the group revealed that they produced a trail of seven robotic units.

Speaking for a news release, Bas Overvelde, the Principal Investigator of the Soft Robotic Matter group, stated, "Ultimately, we want to be able to use self-learning systems constructed from simple building blocks, which for example, only consist of a material like a polymer. We would also refer to these as robotic materials."