This article presents the locomotion analysis and optimization of actinomorphic soft robots, which are composed of soft arms actuated by shape memory alloy wires. The soft arm that is a composite modular structure is actuated by a self-sensing feedback control strategy. A theoretical model was established to describe the deformation of the soft arm, combining the Euler–Bernoulli beam model of the soft arm with the constitutive model and the heat transfer model of the shape memory alloy wire. The kinematics of the actinomorphic soft robot was analyzed using the modified Denavit–Hartenberg method, and the motion equation of the actinomorphic soft robot was presented based on the quasi-static hypothesis. Results show that the actinomorphic soft robot moves with a zig-zag pattern. The locomotion of four actinomorphic soft robots with three to six arms was analyzed, and the gait parameters of each locomotion type were optimized. The optimization results indicate that the three-arm actinomorphic robot with certain gait parameters has the best performance and achieves a maximum stride length of 75 mm. A series of experiments were conducted to investigate the movement performance of the three-arm actinomorphic robot in various environments.