Learning to Simulate on Sparse Trajectory Data

Simulation of the real-world traffic can be used to help validate transportation policies. A good simulator means simulated is similar traffic, which often requires dense trajectories (i.e., with a high sampling rate) cover dynamic situations in real world. However, most cases, are sparse, makes simulation challenging. In this paper, we present novel framework ImInGAIL address problem learning simulate driving behavior from sparse data. The proposed architecture incorporates data interpolation process imitation learning. To best our knowledge, first tackle sparsity issue for problems. We investigate on both synthetic and trajectory datasets vehicles, showing that method outperforms various baselines state-of-the-art methods.