Open danfeiX opened 2 years ago
Hi @danfeiX ,
That's possible, at least in theory. The only reason it's not implemented is that we need a zarr buffer to write on while simulating, so that's why in this implementation we limit the unroll to the length of the zarr buffer.
A work around would be to replicate the last frame for example.
Clearly, some of the features will not be available (e.g. traffic lights) and no new agent will appear in the scene
Hi there,
Thanks for the great work and codebase!
I have a question regarding rolling out a simulation model. With both the ego and agents controlled by learned models, in theory the rollout episode length should not be limited by the length of the scene dataset. I.e., one should be able to run the simulation for longer than the number of frames in that scene. Is there a reason for limiting the episode length or is it just not implemented?
Thanks.