Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 14: Machine Learning in Dynamics and Statistical Physics II
DY 14.7: Talk
Monday, March 9, 2026, 16:30–16:45, HÜL/S186
Physical embodiment enabled learning for autonomous navigation of active particles in complex flow fields — •Diptabrata Paul1, Nikola Milosevic2, Nico Scherf2, and Frank Cichos1 — 1Molecular Nanophotonics Group, Peter Debye Institute for Soft Matter Physics, Leipzig University, 04103 Leipzig, Germany — 2Max Planck Institute for Human Cognitive and Brain Sciences, 04103 Leipzig, Germany
Autonomous navigation at the microscale is a major challenge in active matter due to strong environmental noise and hydrodynamic disturbances. While living systems rely on sophisticated sensing and feedback to regulate functions from sub-cellular processes to chemotactic navigation strategies, artificial microswimmers lack such adaptive mechanisms and therefore struggle to respond effectively to stationary or dynamic perturbations. In this work, we introduce an actor-critic reinforcement learning (RL) framework and demonstrate that physical embodiment alone enables adaptive navigation without explicit environment sensing. Training of the active particle agent in strong and spatially varying flow fields leads to emergence of robust strategies that counteract hidden hydrodynamic perturbations excluded from the agent’s observation space. This reveals that embodied dynamics encode sufficient information for effective decision-making, enabling RL to exploit morphology-environment coupling as an implicit sensing channel. Our approach bridges the gap between simple stimulus response schemes and higher-level adaptive behavior and establishes a foundation for online learning, and microscale robotics.
Keywords: Active Matter; Microswimmer; Reinforcement Learning; Intelligent systems