Multi-Modal Simultaneous Forecasting of Vehicle Position Sequences using Social Attention