Maze-solving in a plasma system based on functional analogies to reinforcement-learning model.

Area of Science:

• Plasma Physics
• Computational Science
• Analog Computing

Background:

• Maze-solving is a classical problem with recent analog solutions using diverse media.
• Plasma generation in confined channels offers a novel medium for route-finding applications.

Purpose of the Study:

• To experimentally investigate maze-solving capabilities of plasma in a channel network.
• To computationally generalize plasma route-finding behavior using a reinforcement-learning model.

Main Methods:

• Experimental setup: Mixed direct-current (DC) volume and alternative-current (AC) surface dielectric-barrier discharge in a narrow channel labyrinth (1x1 mm cross-section).
• Plasma ignition and route observation in low-pressure Argon gas, with electrodes at entry and exit.
• Computational modeling using a reinforcement-learning approach, mapping electric potentials to rewards.

Main Results:

• Observed plasma successfully navigating from entry to exit in the channel maze as discharge voltage increased.
• Quantified route convergence using Shannon entropy, demonstrating effective maze-solving.
• Reinforcement-learning model reproduced similar short-path routes, validating the analog computing function.

Conclusions:

• Plasma-channel networks exhibit analog computing functions analogous to modified reinforcement-learning algorithms.
• This study establishes plasma as a viable medium for complex route-finding tasks.
• Findings suggest potential for novel computing paradigms based on plasma dynamics.