Chemotactic navigation of a multi-link microrobot using reset-free hierarchical reinforcement learning

In this talk, we will demonstrate chemotactic navigation of a multi-link microrobot using hierarchical reinforcement learning (RL). RL enables the robot—with chain or ring topology—to acquire topology-specific swimming gaits like flagella-like wave propagation or amoeboid oscillations. These microswimmers navigate chemotactically in biologically relevant scenarios, including conflicting chemoattractants, pursuing a bacterial mimic, steering in vortical flows, and squeezing through constrictions. We also achieve reset-free, partially observable RL, addressing challenges of manual resets and partial observability in real-world microrobotic RL.