We'd like to train and evaluate the simulated Locobot on object navigation in ProcThor homes as reported in the ProcThor paper. The infrastructure for this seems to be split into different GitHub repositories. Could you help us understand which pieces we need from which repository to replicate your experimental setup?
Given my current understanding, I think the procthor repository only generates homes but doesn't simulate them. So we could either generate homes using procthor or use the pregenerated procthor-10k dataset in prior. We would then somehow load those houses into ai2thor and combine them with the locobot from robothor-challenge, and for evaluation use the object-nav-eval dataset from prior?
We'd like to train and evaluate the simulated Locobot on object navigation in ProcThor homes as reported in the ProcThor paper. The infrastructure for this seems to be split into different GitHub repositories. Could you help us understand which pieces we need from which repository to replicate your experimental setup?
Given my current understanding, I think the procthor repository only generates homes but doesn't simulate them. So we could either generate homes using procthor or use the pregenerated
procthor-10k
dataset in prior. We would then somehow load those houses into ai2thor and combine them with the locobot from robothor-challenge, and for evaluation use theobject-nav-eval
dataset from prior?Thanks a lot for your help!