This paper introduces ASALT, a method that enables lateral transfer learning in multi-agent reinforcement learning by accommodating mismatched state-space dimensionalities between source and target domains. The approach uses observation-level and state-level adapters to map inputs into a shared embedding space, facilitating effective knowledge transfer across heterogeneous environments.
- ASALT incorporates observation-level and state-level adapters to map target-domain observations and global states into a shared embedding space.
- Experimental results show that ASALT surpasses existing baselines in sample efficiency and global return in cooperative settings.
- The method mitigates negative transfer, a major obstacle when transferring policies between domains with differing observation and action spaces.
ASALT allows for more effective strategy transfer across heterogeneous domains where previous methods required identical dimensionalities, thereby addressing the challenge of negative transfer in multi-agent systems.