Surprise Potential as a Measure of Interactivity
in Driving Scenarios

Under Review

In this paper, we present a novel metric that identifies interactive scenarios by measuring an AV's surprise potential on others. First, we identify three dimensions of the design space to describe a family of surprise potential measures. Second, we exhaustively evaluate and compare different instantiations of the surprise potential measure within this design space on the nuScenes dataset. To determine how well a surprise potential measure correctly identifies an interactive scenario, we use a reward model learned from human preferences to assess alignment with human intuition. Our proposed surprise potential, arising from this exhaustive comparative study, achieves a correlation of more than 0.82 with the human-aligned reward function, outperforming existing approaches.

Distribution of Surprise Potential of Different Datasets

nuScenes training split

nuScenes validation split

              

nuPlan-mini training split

nuPlan-mini validation split

              

Waymo training split

Waymo validation split

              

Download Interactivity Score (To be updated)

You can download the interacivity score of the datasets used in the paper from the following links. Note that the score is for each segment of the dataset, which is obtained from the trajdata dataloader. Each segment has 5 second history and 4 second future with timestep interval of 0.5 second. Each file contains a dictionary with the following keys:

  • idx: The list of the name of the scene.
  • ts: The list of the start timestep of the future.
  • data_idx: The list of the index of the data sample in the trajdata dataloader.
  • score: The list of the (unnormalized) interactivity score of the segment.
Dataset Split Segment number Download Link
nuScenes Train 15530 Google Drive
nuScenes Validation 3319 Google Drive
Waymo Train 487002 Google Drive
Waymo Validation 44097 Google Drive
nuPlan mini Train 19387 Google Drive
nuPlan mini Validation 4074 Google Drive

BibTeX

        @article{ding2025surprise,
          title={Surprise Potential as a Measure of Interactivity in Driving Scenarios},
          author={Ding, Wenhao and Veer, Sushant and Leung, Karen and Cao, Yulong and Pavone, Marco},
          journal={arXiv preprint arXiv:2502.05677},
          year={2025}
        }
Template from Nerfies