[MORL Generalization] Collection of useful code fixes and improvements from our MORL Generalization paper

[JaydenTeoh]

Hey Florian and Lucas,

My paper On Generalization Across Environments In Multi-Objective Reinforcement Learning with Peter Vamplew is out now. Thank you for the assistance regarding the code over discord the past few months. Here are some code fixes and improvements I collated while working on the MORL-Generalization benchmark:

• Add MOSAC Discrete, weights saving+loading for MORL-D, net_arch specification for MOSAC #134
• Add PGMORL 3D and include sparsity coefficient from paper #135
• Add proper termination functions for GPI-PD LunarLander and Humanoid #136
• fix mario wrappers not compatible with gymnasium==1.0.0 and fix mario video recording #137
• Add MOAsyncVectorEnv wrapper Farama-Foundation/MO-Gymnasium#115
• Fix mario video not recorded with standard gymnasium RecordVideo Farama-Foundation/MO-Gymnasium#116

I will look through my codebase and see if there are any more things to add on in the near future. Apologies for dumping so many PRs all at once.

Our MORL-Generalization codebase can be found here - https://github.com/JaydenTeoh/MORL-Generalization.

[LucasAlegre]

Hi @JaydenTeoh!

Thank you very much for all these contributions! It's awesome to see the community building upon our work :)

I will be at a conference next week, but I'll try to review these the following week.

[ffelten]

Hey @JaydenTeoh,

I've made a first pass on some of the PRs. Overall, the code looks good and will be valuable for the community.
I essentially saw minor things and pre-commit related stuff--which should be fixed in a matter of seconds.

Thank you 🙏.