instadeepai / Mava

🦁 A research-friendly codebase for fast experimentation of multi-agent reinforcement learning in JAX
Apache License 2.0
737 stars 90 forks source link

Chore: Connector Update #1077

Closed SimonDuToit closed 4 months ago

SimonDuToit commented 6 months ago

Update to Connector wrapper and scenarios to bring it in line with the version used in benchmarks. Previously, other agents and targets were marked with binary variables. This has been replaced with integers, so that agents can be distinguished from each other and matched to their corresponding target. Also, rewards and discounts are shared between agents, as this gave the best performance.

RuanJohn commented 4 months ago

Thank you for this @SimonDuToit! Something that would be really nice is if we could add the vector obs versions of cleaner and connector in Mava as well.

sash-a commented 4 months ago

Thank you for this @SimonDuToit! Something that would be really nice is if we could add the vector obs versions of cleaner and connector in Mava as well.

Can we do this in a separate PR though?