Pneumatic muscle actuators have the highest power/weight ratio of any actuator and therefore it has a broad application prospect in soft robotics. This paper presents a two degree of freedom (2-DOF) pneumatic muscle actuator that consists of four pneumatic artificial muscles (PAMs) and two rotation joins. However, this pneumatic muscle actuator has highly nonlinear and hysteretic properties (among muscle force, muscle displacement and pressure in the muscle), which lead to difficulty in accurate position control. Pneumatic muscle actuator usually necessitate the use various nonlinear techniques for control in order to improve their performance. The paper contains selection of computational intelligence methods designed for application in pneumatic muscle actuator. There are described and compared MLP neural network and RBF neural network.