Design and Preliminary Application Study of an AI Driven Personalized VR Rehabilitation Support System for Children with Autism

Abstract.Objective To address insufficient task adaptation, fragmented process information, and a lack of transparency in the basis for training adjustments for children with autism, this study designs an AI driven personalized VR rehabilitation support system and establishes a preliminary application evaluation framework. Methods Drawing on TEACCH structured teaching to organize tasks, the system integrates the DeepSeek large language model, Unity virtual environments, and MQTT communication to support ability profiling, strategy recommendation, training interaction, and coordination across multiple terminals. Candidate strategy screening, output validation, and therapist confirmation form a closed loop for task adjustment. Technical system tests and a small scale feasibility study are used to evaluate operational performance, training acceptability, and task performance. Results A system design covering perceptual, communication, and social tasks was developed, together with four functional modules—Explore at Your Own Pace, Fun Practice Zone, Everyday Scenarios Town, and Peer Play Community—and corresponding evaluation indicators. Conclusion This study proposes a VR rehabilitation support system that combines structured tasks, AI recommendations, and professional supervision. Actual system feasibility and training related outcomes must be assessed using application data; no definitive conclusion about rehabilitation effectiveness can yet be drawn.