Abstract: Conventional space rigid-body robots are subject to defects such as limited degree of freedom and low environmental adaptability. For those problems, a soft robot based on the organism structure is designed inspired by inchworm and snake, which is applicable to space in-orbit service. Firstly, the prototype of the soft robot is built and the drive characteristics of the Ni-Ti shape memory alloy (SMA) as the actuators are studied. A visual control interface is designed and the manoeuvrability of the soft robot prototype is verified by physical experiments. Besides, the Q-learning algorithm and the corresponding reward function based on the designed soft robot structure are designed. A simulation model of the soft robot is built, and autonomous learning and planning experiment of the robot arm model based on Q-learning is completed in the simulation environment. The experiment results show that the robot arm can converge to a stable state in a short time and complete planning tasks independently. Therefore, the proposed algorithm is effective and feasible, and the reinforcement learning method has a good application prospect in the intelligent planning and control of soft robots.