RL-MINDS: Reinforcement Learning For Mobility-Induced Duty-Cycles In WSN For Underground Mines
Abstract
Underground mines increasingly depend on pillarmounted wireless sensor networks to deliver safety-critical measurements-temperature, humidity, toxic-gas levels, and mobility cues-to miners' mobile devices, which opportunistically relay data during encounters. However, underground environments impose severe energy constraints: battery replacement is hazardous and costly, energy harvesting is scarce, and mobile charging access is limited, leaving many areas underserved. As a result, sensors must operate with extreme energy discipline while maintaining timely and reliable data delivery. Existing duty-cycling and scheduling methods often rely on static policies, require manual penalty tuning, or fail to adapt to dynamic miner mobility, spatially varying channel conditions, and heterogeneous battery states. To overcome these limitations, we propose RL-MINDS, a constrained reinforcement-learning (RL) framework that formulates duty-cycle scheduling as a constrained Markov decision process (CMDP) to jointly optimize energy consumption and net bit rate. RL-MINDS enforces energy and battery constraints using adaptive Lagrangian multipliers updated through proportional-integral-derivative (PID) control, eliminating manual tuning and ensuring stable constraint satisfaction. Meanwhile, the RL agent selects sensor duty cycles conditioned on spatial variations in miner density, underground channel conditions, and individual battery states. An attention-augmented variational autoencoder further compresses high-dimensional network states, enabling scalable policy learning. Simulations on a realistic 100sensor underground network show that RL-MINDS achieves a 28% higher net bit rate than the strongest baseline, utilizes 95% of the energy budget, and reduces energy-constraint violations to 1.4% at convergence.
Recommended Citation
M. A. Yadav et al., "RL-MINDS: Reinforcement Learning For Mobility-Induced Duty-Cycles In WSN For Underground Mines," Proceedings IEEE International Conference on Mobile Data Management, pp. 1 - 11, Institute of Electrical and Electronics Engineers, Jan 2026.
The definitive version is available at https://doi.org/10.1109/MDM71479.2026.00012
Department(s)
Computer Science
Second Department
Mining Engineering
Keywords and Phrases
duty-cycle; proximal policy optimization; underground mines; variational autoencoder; Wireless sensors
International Standard Serial Number (ISSN)
1551-6245
Document Type
Article - Conference proceedings
Document Version
Citation
File Type
text
Language(s)
English
Rights
© 2026 Institute of Electrical and Electronics Engineers, All rights reserved.
Publication Date
01 Jan 2026
Comments
Centers for Disease Control and Prevention, Grant None