Abstract: Intelligent robots are highly integrated complex electromechanical system that incorporates electromechanical transmission, motion control, environmental perception, and artificial intelligence. At present, the performance of the core components such as driving motor, sensor and dexterous hand is insufficient, which leads to the superposition of mechanical error, positioning error, and sensor error in intelligent robot. Besides, with a lacking of fixed motion base, intelligent robots have large execution error in practical tasks. Also the dynamic performance of motion planning and control algorithm is not sufficient, the generalization ability of the large model is limited, and the capacity and effectiveness of robot training dataset are not up to standard. These technical bottlenecks lead to low stability and reliability of intelligent robots in task execution and limited industry applications. Many research institutions and researchers believe that intelligent robots can enhance their stability and generalization ability through training. And the leading enterprises in the robotic industry have also begun to lay out robot training platforms to accelerate the application of intelligent robots. Based on the technological progress and development trends in the intelligent robot training field, firstly, the purpose of intelligent robot training is expounded. Then, several current main simulation training platforms and real training platforms for intelligent robots are introduced, and the advantages and disadvantages of the simulation training platforms and real training platforms are further analyzed. Finally, for the training of intelligent robots, an idea of "simulation + reality" comprehensive training based on the division of functional modules is proposed, hoping to provide reference and inspiration for subsequent research.