Downloads: 1
India | Computer Science and Engineering | Volume 14 Issue 10, October 2025 | Pages: 425 - 429
A Hybrid Deep Reinforcement and Fuzzy Logic Framework for Adaptive Decision Making in Dynamic Environments
Abstract: Dynamic, partially observable, and stochastic environments require decision-making systems that combine learning flexibility with robust, interpretable reasoning. This paper proposes a hybrid framework that integrates Deep Reinforcement Learning (DRL) with Fuzzy Logic (FL) to produce adaptive policies that are both performant and interpretable. The hybrid architecture uses a DRL module (actor-critic family) for representation learning and long-term optimization, while a fuzzy reasoning module provides high-level rule-based adjustments, safety constraints, and interpretability. We detail the framework architecture, learning algorithm, experimental setup across simulated dynamic tasks (navigation with changing goals, resource allocation with fluctuating demand, and nonstationary control with drifting dynamics), and evaluation metrics. Results show that the hybrid system improves sample efficiency, reduces catastrophic failures under distribution shift, and provides human-readable decision rationales compared to baseline DRL agents.
Keywords: Deep Reinforcement Learning, Fuzzy Logic, Adaptive Decision Making, Hybrid Framework, Dynamic Environments, Policy Adaptation, Robust Control
How to Cite?: Dr. M. V. Siva Prasad, Dr V. Subrahmanyam, "A Hybrid Deep Reinforcement and Fuzzy Logic Framework for Adaptive Decision Making in Dynamic Environments", Volume 14 Issue 10, October 2025, International Journal of Science and Research (IJSR), Pages: 425-429, https://www.ijsr.net/getabstract.php?paperid=SR251006184000, DOI: https://dx.doi.org/10.21275/SR251006184000