Researchers from Santa Clara University, the New Jersey Institute of Technology, and the University of Hong Kong have used deep reinforcement learning to teach microrobots how to swim. This is a big step forward in the field of microswimming. The creation of artificial microswimmers that can travel the globe in a manner akin to naturally occurring swimming microorganisms, including bacteria, has generated a great deal of attention. These microswimmers hold promise for a wide range of upcoming biomedical applications, including microsurgery and tailored medication administration. But most of the artificial microswimmers on the market today can only move in a small number of fixed ways.
The researchers reasoned that microswimmers may learn and adapt to changing conditions through AI in their work, which was published in Communications Physics. Like people who are learning to swim, microswimmers need reinforcement learning and feedback to stay afloat and move in different directions when conditions change. However, the physics of the tiny environment gives them their own set of challenges.
According to On Shun Pak, an associate professor of mechanical engineering at Santa Clara University, “being able to swim at the micro-scale by itself is a tough issue.” The design of an artificial microswimmers locomotory gaits can easily become impenetrable when you want it to do more complex moves.
The team was able to successfully train a basic microswimmer to swim and navigate in any direction by fusing reinforcement learning and artificial neural networks. The swimmer receives input on how effective specific movements are when they do certain movements. The swimmer then slowly figures out how to swim by interacting with its surroundings.
According to Alan Tsang, assistant professor of mechanical engineering at the University of Hong Kong, “the microswimmer learns how to manipulate its ‘body parts’ — in this example, three microparticles and extensible connections — to self-propel and turn.” Artificial microswimmers accomplishes this by using solely a machine learning algorithm, not human understanding.
The AI-powered swimmer can move in any direction by switching between different ways of moving on its own. The researchers used the swimmer’s impressive abilities to prove that it could follow a complex path without being expressly trained. They also looked at how well the swimmer navigated while being affected by fluid flows from the outside.
Professor of mathematical sciences at New Jersey Institute of Technology, Yuan-nan Young, said, “This is our first step in facing the task of generating microswimmers that can adapt like biological cells in navigating complex surroundings autonomously.”
Artificial microswimmers will need to be able to adapt to their environment if they are to be used in the future in environments that are hard to control and full of surprises.
According to Arnold Mathijssen, a University of Pennsylvania expert on microrobots and biophysics who was not involved in the study, “This work is a key example of how the rapid development of artificial intelligence may be exploited to tackle unresolved challenges in locomotion problems in fluid dynamics.” By combining machine learning and microswimmers in this work, more connections will be made between these two hotly debated areas of research.
