google-research/relay-policy-learning

Compound goals data

etaoxing opened this issue · 2 comments

The demo data seems to be separated into different tasks. Were the VR demo users given a desired goal for each task, and is this goal data available?

To ensure that the data distribution is uniform across various elements in the scene, users were given general guidelines for which subset of elements to interact with. A random subset was selected for each new interaction session. The data from all the sessions were merged together (without any session metadata) to ensure interaction uniformity and session anonymity.

Ok thanks! It seems like d4rl has integrated the environment with the object goals.