Request: Support for vectorised environments #253

[AsadJeewa]

Hi All

Is it possible to run DMLab2d with vectorised environments? I am using CleanRL + Shimmy to run MeltingPot but when I try to use concat_vec_envs_v1 I get error: TypeError: cannot pickle dmlab2d.dmlab2d_pybind.Lab2d object. Is there any other way? I need to train faster as I have limited compute walltime and the easiest way is vectorised environments. Any other methods for optimisation?

[jduquevan]

This is also an issue when running Meltingpot + torchrl or Meltingpot + gymnasium for parallelization. Adding support to pickle Lab2d objects would help resolve other issues e.g.: meltingpot, torchrl .

[tkoeppe]

@charlesbeattie: could you maybe have a look?

[AsadJeewa]

Are ther any updates on this?

[charlesbeattie]

Would it be possible to pickle the settings and construct the environment where needed?