You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi Qiang,
the challenge is to learn one policy from each dataset. Learning from multiple datasets is not allowed. The pre-stage datasets are consequently not to be used in the real-robot stage.
The idea behind this rule is that we want to see how a submission deals with each dataset separately. The real-robot data is more challenging because there is noise on the observations and because of the small differences between the robots. Additionally the mixed datasets test whether an algorithm gets distracted by suboptimal trajectories. We think it's important for offline RL to be able to learn from imperfect data as in real-world applications high-quality simulators are often not available and the collected data might be from several sources of varying quality.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Hi,
May I know if it is possiable to use the pre-stage dataset in the real robot stage? Thank you.
Kind regards,
Qiang
All reactions