Avnish Narayan

Just looked into this. We moved `TFTrainer` to the base of garage, so you can now import it by calling `from garage import TFTrainer` Sorry for the slow replies. At...

I think I speak for all the maintainers when saying we definitely would love to do this, but we don't have the time. If there is anyone who is inclined...

Yeah that's probably because there's a lot of information that a rnn like the one in rl2 captures dynamics between timesteps. We tried to overcome this by stacking consecutive timesteps...

Yeah PEARL in general we found didn't scale well to complex environments and there are predecessors to the algorithm that were published. It was also very slow (the experiments when...

To be honest, I never tried rendering the environments on a gpu instance. I would run them headless and then render on my local non gpu machine, as I didn't...

We've had baselines available in our sister repo, garage, for a while now. I'm updating them with what I ran to get the v2 baseline results. You can track my...

I'll have to get back to you on that @luckeciano. Although I'd recommend re-running your results for any benchmarks that you decide to use, because there might be algorithmic differences...

It seems that people have had a lot of interest in running mt1 examples. We didn't publish the results for these, as we didn't have the time to run results...

Oh those are not the files you want to use then to get those results. What you're looking for instead is ppo_metaworld.py and sac_metaworld.py. On that same branch.

Hey Neeraj, I can spend a little time this week updating the mujoco version. I think the short answer to your problem is that there is a version of mujoco-py...