FPGA Implementation of Real-Time Forward Kinematics for 6-DOF Parallel Manipulators

In recent years, the Stewart platform has been widely adopted in high-precision applications such as robotic surgery and industrial automation, yet solving its forward kinematics problem in real time remains challenging due to computational complexity. Existing numerical and learning-based methods often face trade-offs between accuracy and speed. This paper presents a novel forward kinematics solution method that leverages neural network to provide initial pose estimates, which are then refined through an optimized iterative process. First, a dataset is obtained by truncating a Gaussian distribution and used to train a neural network, which fits the pose as the initial value for the Newton-Raphson iteration, accelerating the convergence process. Second, a simulation model is established to evaluate the accuracy of the proposed method. To enable real-time performance, the proposed method is implemented on a field-programmable gate array (FPGA) with hardware optimizations, including precomputed Jacobian matrices and efficient trigonometric calculations. Experimental results demonstrate that the proposed method achieves the pose error below 0.01 cm in position and 0.01 deg in orientation, with a computation time of just 0.2467 ms per solution. The FPGA implementation maintains high precision while meeting the stringent timing requirements of industrial applications, offering a practical advancement for Stewart platform control systems.