Core theoretical motivations:
1. Generalisability of MARL settings:
1. [[Human Timescale Adaptation]]
2. [[Original XLand Paper]]
2. Social RL leads to interesting insights:
1. [[Jacques et al 2018: Social Influence as Intrinsic Motivation for Multi-Agent Deep Reinforcement Learning]]
2.
3. Environments:
1. [[XlandMinigrid, XLand-1b]]
2. [[Craftax]]
3. [[SMAC]], and now also [[SMAX]] as part of JAX-MARL
4. Roles and coordination in MARL:
1. Specific Algorithms:
1. [[ROMA]]
2. [[SIRD]]
3. [[Dynamic role discovery and assignment inmulti-agent task decomposition]]
4.
2. Environments/benchmarks that current papers are using:
1. [[SMAC Micromanagement]]
2. [[Custom Grid-based benchmarks]]
5. Unsupervised Environment Design
1. [[JAX-UED library by Foerester]]
2. [[Craftax integration]]
3. Specific UED techniques:
1. [[PAIRED, Jacques et al]]
2. [[Prioritized Level Replay]]
6. Cognitive Science Angle:
1. [[One Hour, One Life]]
2. [[The Secret to our Success book]]
7. Misc:
1. [[Meta-Reinforcement Learning Robust to Distributional Shift Via Performing Lifelong In-Context Learning_]]