Efficient multi-user wireless collaborative Virtual Reality (VR) systems are becoming more and more widely used in the field of education, especially during the epidemic period, teachers and students can teach and watch movies together through VR devices. However, the following challenges need to be addressed: 1. How to achieve efficient VR content delivery and high-quality user experience under limited wireless network resources? 2. How can we efficiently realize the scenario of multiple users watching the same VR video? To address these challenges, this paper introduces a novel wireless collaborative VR system. It proposes a 360-degree video transmission algorithm based on motion prediction, and an adaptive multicast algorithm using a hybrid clustering method to improve transmission efficiency and user experience. Specifically, we apply a Long Short-Term Memory Network (LSTM) model using the real-time position of the primary user’s head to achieve efficient motion prediction, and only transmit the visual image data to the other following users, thus improving the utilization efficiency of network resources. Additionally, in the adaptive multicast algorithm, based on real-time network information, a clustering approach using the Self-Organizing Map (SOM) network and the k-means method realizes the effective partition of multicast groups. Then, VR content quality selection is carried out according to the status of the multicast group network, further reducing the transmission content within the system. We experimentally verified the accuracy of the LSTM model in achieving motion prediction, and verified our algorithm by deploying it to commodity mobile devices. Our system improves the quality of experience (QoE) index by nearly 74% and the frame rate index by more than 30% compared to the state-of-the-art (SOTA) technology.