Aleksei Petrenko

@edbeeching I added a few changes to fix that Atari-only flag in enjoy.py while keeping the optional "smooth rendering" functionality for other envs like VizDoom. I believe this should not...

@edbeeching I made some more changes so that tests actually fail instead of freezing if there's an exception in the environment. Also removed [dev_envpool] in setup.py because I believe it's...

I'd be interested in that too. Basically this means rendering multiple viewpoints in the same environment within the same engine. This is typically not what engines are designed for, but...

Or you can just switch to VizDoom! https://github.com/mwydmuch/ViZDoom/issues/391

Thanks for feedback @charlesbeattie Another option for anyone looking: https://github.com/alex-petrenko/megaverse Multi-agent immersive environments 70 times faster than DMLab! (shameless self-promotion)

I think 0.35 seconds is about as fast as it gets. Internally, I believe they restart/reload the level from scratch on every reset. If you're using something like random maze...

There must be a way but I'm not an expert. I'd guess you can just do it on the level of the scenario, basically running the same very long episode...

Sorry, I'm relatively new too. My first suggestion is to measure how much you're actually losing. Try to minimize the losses by making your episode longer and/or doing all resets...

The simplest thing I can think of is to just make one of the actor workers a "special" one. I.e. add some sort of flag that will also save environment...

@signalprime In case the evaluation worker has a different set of environments and maybe an entirely different workflow to the regular actor worker, I'd say you need to create/spawn a...