About score calculation

MCZhi / DIPP

[TNNLS] Differentiable Integrated Prediction and Planning Framework for Urban Autonomous Driving

https://mczhi.github.io/DIPP/

209 stars 41 forks source link

About score calculation #27

Closed wangyy0 closed 10 months ago

wangyy0 commented 10 months ago

Hi, thank you for sharing the great work!

While working on GameFormer, I noticed that you used a Gaussian distribution, which is not used in DIPP. Do you think adding Gaussian distribution will be more beneficial for trajectory scoring and subsequent training?

Bests.

MCZhi commented 10 months ago

Hi, @wangyy0. Sorry for the late reply. Yeah, I believe adding Gaussian distribution would be helpful in trajectory scoring but would not bring significant improvement. It may also make the training unstable somehow, so be careful with that.

wangyy0 commented 10 months ago

thank you!