<p>The double-chain space manipulator (DCSM) can complete collaborative tasks in a large workspace, which is of great significance for its development. The complex structure and multiple variables of the DCSM present significant challenges for DCSM design. In this paper, an integrated type and dimension method for DCSM design of using a Q-learning-based mixed-integer optimization method was proposed. Based on the analysis of the mechanism characteristics of the DCSM, a model-free kinematics modeling method was proposed for unknown configurations, and the discrete variables, including the number and axis direction of joints, and the continuous variables, including the link lengths, were linearized, enabling the subsequent efficient optimization. Then, a performance index system, including workspace, comprehensive operability and follow-up sensitivity, was established, which reflects the coupling relationship between the main chain and the branch chains with regard to performance. By introducing the ideas of judgment and decision-making from Q-learning into the mechanism design, efficient optimization of multiple variables under complex performance constraints was achieved. The analysis results indicate that the method proposed in this paper has high convergence speed and computational efficiency, and can obtain multiple feasible solutions of different types. This study provides the basis for the design of manipulators with complex configurations and multiple variables.</p>