about offline RL about social-nce HOT 1 CLOSED

Hi there,

We use the CrowdNav benchmark to set up the RL task. We collected the data for offline RL when training our rainbow DQN.

The key steps to collect data are:
(i) 10k episodes are collected during online RL training from scratch, (ii) multiple free explorations are carried out using online RL checkpoint models trained after K episodes, where K ∈ {500, 1000, 3000, 5000}.
The offline RL data was collected from the training or simulations in CrowdNav environment. We believe this example showcases the general applicability of our proposed constrastive representation learning technique.

from social-nce.