As robots play more critical roles in diverse and complex scenarios in the real world, monomorphic robots are limited to repeating and rather simple tasks. How to achieve a robust, flexible, and scalable multi-robot system becomes essential research. Model-driven software development (MDSD) provides a sturdy methodology for robotic programming using multilevel domain-specific languages (DSLs). These DSLs lay a solid foundation for the design, integration, and extensibility of robotic applications. In this paper, we propose a deep domain-specific model framework for the self-reproducing robotic control system to escort reliable, versatile tasks of heterogeneous robots.