A Fuzzy-TD3 Hybrid Reinforcement Learning Framework for Robust Trajectory Tracking of the Mitsubishi RV-2AJ Robotic Arm

This paper presents a novel hybrid control architecture that fully integrates fuzzy logic with the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm for advanced trajectory tracking control of a 5-degree-of-freedom (DOF) Mitsubishi RV-2AJ robotic arm. The proposed framework synergistically combines the interpretability, rapid response, and rule-based reasoning of fuzzy logic systems with the model-free, adaptive learning capabilities of deep reinforcement learning. In this architecture, a fuzzy logic supervisor provides immediate compensatory actions based on real-time analysis of tracking error and its derivative, while the TD3 agent concurrently learns optimal continuous control policies to handle complex nonlinear dynamics and long-term performance optimization. Extensive simulation studies conducted across multiple challenging 3D trajectories (N-shaped, helical, and spiral) demonstrate that the hybrid fuzzy-TD3 controller achieves superior tracking performance, with error reductions of 27.8–50% compared to standalone TD3 and 14.8–28.6% over conventional PID-TD3 hybridization. Comprehensive robustness evaluations under parametric uncertainties, payload variations, and external disturbances confirm enhanced disturbance rejection capabilities and stable operation across all performance metrics. The controller's effectiveness is further validated through sensitivity analysis, empirical Lyapunov stability verification, and explainable rule activation patterns, establishing a new benchmark for transparent, adaptive, and high-performance robotic control in industrial automation applications.