To address the issues of difficult parameter tuning and poor dynamic adaptability in traditional Proportional-Integral-Derivative(PID) controllers for robotic manipulators, this paper proposes a Backpropagation (BP) neural network-based PID controller optimization method which innovatively integrates BP neural network with PID to enable real-time parameter adjustment. By leveraging the self-learning capability of BP neural networks to achieve online optimization of PID parameters (proportional, integral, and derivative coefficients), a closed-loop control system is constructed for joints of robotic manipulators. Furthermore, we incorporate system error into the neural network input layer to enhance error sensitivity, and introduce a momentum factor to improve the gradient descent algorithm, which significantly accelerates network convergence. In trajectory tracking simulation experiments, a six-degree-of-freedom (6-DOF) robotic manipulator dynamic model is established in MATLAB/Simulink environment. The results demonstrate that compared with conventional PID controllers, the BP-PID controller achieves 47.2% average reduction in steady-state error, 31.7% decrease in overshoot, and 5.1% shorter dynamic response time during step response, validating the method’s effectiveness and robustness in nonlinear time-varying robotic manipulator system.