A new arXiv paper argues that autonomous driving systems need more than object perception: they must also understand how surrounding objects move and interact over time. The authors warn that missing this dynamic understanding can lead to dangerous decisions on the road.

The paper introduces STRIDE, an evaluation approach for vision-language models (VLMs) that focuses on spatiotemporal reasoning in driving edge cases. Rather than testing only what objects are present, STRIDE appears designed to probe whether models grasp how those objects behave and relate across time.

Because the source is a single preprint with a truncated abstract, the specific design of STRIDE is not detailed here. Still, the motivation is clear: robust driving AI must reason about motion and interaction, not just static perception.