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

Area of Science:

• Robotics
• Artificial Intelligence
• Biomimicry

Background:

• Effective navigation for microswimmers, both biological and artificial, is crucial for target localization using limited environmental cues.
• Biological microswimmers exhibit sophisticated mapless navigation strategies, a capability that remains challenging to replicate in artificial systems.
• Current artificial microswimmers often rely on pre-existing maps, limiting their adaptability in unknown or dynamic environments.

Purpose of the Study:

• To develop an autonomous navigation system for artificial microswimmers capable of searching for targets in unknown environments without pre-existing maps.
• To enable artificial microswimmers to respond to local environmental cues for effective navigation, similar to biological counterparts.

Main Methods:

• Deep reinforcement learning was utilized to train a reconfigurable artificial microswimmer for mapless navigation.
• The microswimmer was trained to navigate towards a chemical source by sensing and responding to local chemical signals.

Main Results:

• The artificial microswimmer successfully navigated towards a chemical source by adapting its locomotion strategy, exhibiting a 'run-and-tumble' behavior akin to bacterial chemotaxis.
• The mapless microswimmer demonstrated robust performance in environments with significant deviations from its training conditions, including fluctuating and time-varying chemical fields.
• The system showed capability in exploring complex chemical landscapes with multiple concentration maxima, efficiently locating target areas.

Conclusions:

• Deep reinforcement learning provides a viable method for creating autonomous, mapless navigation in artificial microswimmers.
• The developed strategy allows microswimmers to effectively search for targets in unknown and dynamic environments, mimicking biological navigation.
• This research paves the way for advanced autonomous microswimmers applicable in diverse, unpredictable settings.