Small group activities have many educational advantages. In this contribution, we present a novel system for monitoring and analysing real-time group dynamics in a video-conferecing environment, with and without an autonomous virtual robot facilitator, which was programmed with group facilitation best practices. Groups of tenth-grade students interacted with the system, where we examined subjective perceptions of the activity, group dynamics and the facilitator, as well as analysed real-time speaker and facial expression recognition data. Our results show that students reported the activity with the virtual robot facilitator to be significantly more efficient, more understandable and more motivating of participation compared with activities without facilitation. Furthermore, students also spoke more, were happier and were significantly more engaged with virtual robot facilitation. Analyzing subjective and object measures showed significant correlations between the students' perceptions of the group activity and their own participation in it. Finally, it was found that group dynamics, as measured by time-series speaker and facial expression data, can be clustered into groups with similar behavioural characteristics and that facilitation with a virtual robot facilitator changed these dynamics. This study suggests that online group activities, facilitated by an autonomous virtual agent, are a promising tool to investigate group dynamics in educational scenarios.