High number of collisions

[BingXueZhao]

Hi, thank you for your great work!

I followed the training setup and hyperparameters as provided in the repository and paper. Most evaluation metrics are consistent with the results reported in the paper. However, I noticed that the collision count is significantly higher in my experiments.

Could you please help me understand why this might be happening? Are there any additional implementation details or post-processing steps not explicitly mentioned in the paper that might affect the collision results?

Thanks in advance for your help!