VLNCE questions

[mbautistamartin]

Thanks for the incredible effort on putting this dataset together! I was wondering how can I find the continuous trajectory actions/camera poses for each episode. If I look into "xR_VLNCE_v0/train/train_guide.json.gz", each episode has a trajectory_id field. Does this correspond to the keys in "RxR_VLNCE_v0/train/train_guide_gt.json.gz"? Or is it episode_id that corresponds?

In addition, where can I find the camera poses (location and rotation) for each trajectory? There's an "actions" field in "RxR_VLNCE_v0/train/train_guide_gt.json.gz", how do the actions integers map to actions (1 forward, 2 turn left, 3 turn right)? What does the "locations" field mean there? I would really appreciate if you could help understand the field structure a bit better.

Thanks!