A microswimmer navigating turbulent flow must solve two problems simultaneously: exploit the flow to reach its target efficiently, and avoid getting trapped by obstacles. Stagnation points near solid surfaces act as attractors — a swimmer caught in one will orbit indefinitely, wasting energy while making no progress.
The paper on obstacle-aware navigation of smart microswimmers (arXiv: 2603.21223) extends adversarial Q-learning methods to environments containing circular obstacles in forced two-dimensional Navier-Stokes turbulence. The obstacle is introduced via volume penalization, which enforces no-slip and no-flow conditions within the solid region. The Q-learning framework is then augmented with a mechanism that suppresses the tendency of swimmers to approach stagnation points near the obstacle surface.
Smart swimmers trained with this obstacle-aware strategy outperform both naive swimmers (which swim directly toward the target) and surfers (which exploit the flow but ignore obstacles). The advantage is not just in avoiding traps — it's in learning when to detour. The optimal policy sometimes routes the swimmer away from the target to avoid the wake region behind the obstacle, accepting a longer path for a faster arrival.
The through-claim: obstacle avoidance and flow exploitation are not independent skills — they interact through the flow field. The obstacle creates the stagnation points, but it also modifies the surrounding turbulence. A strategy that merely avoids the obstacle misses the flow modifications it creates. A strategy that merely exploits the flow gets trapped. The smart swimmer learns the coupled problem: the obstacle as both threat and flow-modifier.
2603.21223. Microswimmers / reinforcement learning / turbulent flow / obstacle avoidance / Q-learning.