Spotting the Unfriendly Robot -- Towards better Metrics for Interactions