Microrobots may swim thanks to neural networks

Specialists from the New Jersey Institute of Technology and their Chinese colleagues have been able to teach microrobots to move in liquid by adapting to environmental changes while using neural networks.

The scholars spoke about the success in an article published in the Communications Physics journal, according to Russia’s Gazeta.ru.

Floating microrobots can be used, in particular, for medical tasks, such as targeted medicine delivery and microsurgery. However, so far, their manoeuvrability is limited.

Microrobots of three fragments, united by tensile joints, could adapt to difficult conditions and move in the fluid in the right direction, regardless of possible interference, through neural networks and reinforcement learning thanks to the researchers. The learning process was similar to that one when a person learns to swim.

When the microrobot moved in a certain way, it received feedback whether its actions were correct. Then microrobot gradually learned to swim based on the experience of interacting with environment.

After such training, microrobots were able to move along complex trajectories without preliminary programming. Moreover, they were able to reliably adapt in case of disturbances caused by the movements of fluid flows.

Adaptive behaviour is critical for the future use of floating microrobots in complex conditions with uncontrollable and unpredictable environmental factors.