Jannik was working on "Training Microswimmers to Navigate in Complex Environments using Reinforcement Learning" supervised by Samuel Tovey. We congratulate him on this wonderful achievement and invite you to attend the defense.
Abstract: In this thesis, the training of multi-agent reinforcement learning models for micro-robotic navigation in different environments was studied as a proof-of-concept check for potential applications. In increasingly intricate tasks, navigation was assessed in straightforward Brownian motion environments, maze-like capillary structures, constant flow fields, and a blood capillary system.Training success is closely connected to the physical boundaries defined by the chosen parameters as size, maximum swim velocity, and rotational speed but is also affected by environmental parameters like flow speed and the shape of boundary structures. The general trend observed assigns more stable and faster navigation to larger and sufficiently fast agents. In more complex environments, the training success rate decreased. However, for successfully trained models, their application to other simulation settings than their original training environment yielded outstanding results. Thus, the basic requirements for basic navigation of microrobots trained with reinforcement learning are fulfilled.
Best of luck Jannik on your upcoming defense and future endeavors!
