7 Additional Experimental Results and Further Analysis

Neural Information Processing Systems 

In this section, we provide further experimental results and analysis, including ablative analysis, analysis on the selection of edge types and re-encoding gap, as well as additional qualitative results. The descriptions of each model setup are provided in Section 8.2. SG: We show the effectiveness of the distinction of agent node types. The reason is that different types of agents have distinct behavior patterns or feasibility constraints. For example, the trajectories of on-road vehicles are restricted by roadways, traffic rules and physical constraints, while the restrictions on pedestrian behaviors are much fewer.