This study developed and technically validated a “Human-AI Collaborative VR Mental Health Support System” to address limitations in accessibility and real-time responsiveness of traditional psychological counseling resources. The system integrates cutting-edge web technologies and AI models to establish a self-looping closed-loop architecture. The sensing layer utilizes Web Bluetooth to capture real-time heart rate signals and calculates heart rate variability (HRV, RMSSD) as physiological indicators of emotional states. The analysis layer employs a custom emotion inference logic to convert physiological signals into emotional status, automatically triggering alerts when heart rate exceeds 120 BPM. The intervention layer combines Web Audio API and WebXR technologies to instantly switch virtual environments—transforming fear-inducing training scenes into calming settings enhanced with alpha wave music—to achieve real-time emotional regulation.Additionally, the system incorporates a generative AI dialogue model trained on 378,450 psychology-related data entries to provide personalized mental health support. In simulated emotional scenarios, the platform achieved an emotion recognition accuracy of 92% with a response latency under 0.8 seconds, demonstrating strong real-time performance and technical stability. The system shows scalability and clinical translation potential, with applications in school-based mental health monitoring, counseling, and home care settings. The results confirm the feasibility and innovative value of human-AI collaboration in real-time mental health support engineering.