This paper presents a framework designed for robot-assisted education of individuals with Autism Spectrum Disorder (ASD). The framework analyzes a specific type of educational task, known as “shoebox tasks”, and applies this analysis to guide robot-assisted task solving. It uses the visual structure of shoebox tasks to generate correct task representations and to plan actions. To verify the feasibility of automated task analysis, we conducted preliminary experiments using vision-based machine learning and Large Language Models (LLMs).