Driving behavior explanation with multi-level fusion

In this era of active development autonomous vehicles, it becomes crucial to provide driving systems with the capacity explain their decisions. work, we focus on generating high-level explanations as vehicle drives. We present BEEF, for BEhavior Explanation Fusion, a deep architecture which explains behavior trajectory prediction model. Supervised by annotations human decisions justifications, BEEF learns fuse features from multiple levels. Leveraging recent advances in multi-modal fusion literature, is carefully designed model correlations between and mid-level perceptual features. The flexibility efficiency our approach are validated extensive experiments HDD BDD-X datasets.