The work aims to propose an evaluation method of anthropomorphic modeling characteristics of intervention robots based on the emotional experience characteristics of children with ASD to improve the training effect of intervention robots for children with ASD and to provide reference for designers and teaching assistants. Firstly, the anthropomorphism scores of the robot modeling samples in the market were used to select representative samples. Secondly, eight representative indicators were selected from more than 20 HRV indicators and data were collected through HRV experiments. Again, the improved PAD emotion scale was used to obtain the subjective scores of the emotional experience of children with ASD. Then, the HRV that had a greater influence on the emotional experience scores were selected through stepwise regression method principal factors. Finally, the regression relationship between the principal factors of HRV indicators and emotional experience scores was established to provide a theoretical basis for the design and program evaluation of intervention robot modeling for children with ASD. The study showed that geometric robots tended to bring about "relaxed" emotional experiences, humanoid robots brought about "happy, dependent" emotional experiences, and life-like robots might bring about "anxious, dependent" emotional experiences. At the same time, there was a correlation between the pleasantness, arousal, and dominance of children with ASD and the HRV index. Applying the improved PAD emotion scale to the stylistic evaluation of intervention robots for children with ASD and establishing a regression model between emotional experience and HRV indexes provide a theoretical basis for designing and evaluating suitable robots for children with ASD and better promote the development of intervention robots for children with ASD.