The article contains the results of identification of manipulator arm dynamics that is driven by pneumatic artificial muscles (specifically, fluidic muscles). The NARX architecture model - a multilayer perceptron network (MLP model) was selected for identification. Simulations were realized using MATLAB, especially with use the System Identification Toolbox. The simulation results were tracked at 3 different delays for the upper and lower axis of the manipulator separately. The NRMSE criterion (Normal Root Mean Square Error) serves to compare simulation results by comparing output of simulations and measured output data.