Recommended Way to Use NavSim (2 Hz) with Models Expecting 10 Hz Inputs

[pooja-kabra-tri]

Hi — thank you for releasing NavSim!

I have a question regarding temporal resolution.

NavSim scenes are provided at 2 Hz (0.5s intervals), but many existing planning/perception models (including ours) assume 10 Hz (0.1s intervals) inputs. This creates a mismatch between dataset frequency and model expectations.

Current Situation

NavSim data: 2 Hz

Model: assumes fixed 10 Hz temporal spacing

Question

What is the recommended way to handle this mismatch when using NavSim?

Specifically:

Is it acceptable to upsample 2 Hz → 10 Hz using image + ego state interpolation?

We want to align with the intended usage of the dataset and avoid introducing artifacts that could affect evaluation.

Any guidance from the authors would be greatly appreciated.

Thanks!