Learning to Navigate in Chemical Fields Without A Map at Low Reynolds Numbers.

Effective navigation of microswimmers relies on their ability to search for unknown target locations using limited information provided by local environmental cues. Biological microswimmers have evolved versatile strategies to achieve such mapless navigation. Yet, achieving autonomous navigation in artificial microswimmers, comparable to that of their biological counterparts, remains a significant challenge. In this work, deep reinforcement learning is employed to equip a reconfigurable artificial microswimmer with the ability to navigate and search for a target without relying on a pre-existing map. These results demonstrate how this mapless swimmer can effectively navigate toward a chemical source by responding to local chemical signals. Remarkably, the swimmer adapts its locomotory gaits in response to local chemical fields, exhibiting a run-and-tumble strategy reminiscent of bacterial chemotaxis. Unlike map-based swimmers, which depend on pre-existing maps for successful navigation, the mapless swimmer achieves robust performance even in chemical fields that significantly deviate from its training environment, including time-varying fluctuating environment. Moreover, the swimmer can progressively explore complex chemical fields with multiple local maxima, effectively searching for regions of higher concentration. These findings present a promising approach toward achieving autonomous navigation for artificial microswimmers in unknown environments.