STAIRS-Former: Spatio-Temporal Attention with Interleaved Recursive Structure Transformer for Offline Multi-task Multi-agent Reinforcement Learning

STAIRS-Former: Spatio-Temporal Attention with Interleaved Recursive Structure Transformer for Offline Multi-task Multi-agent Reinforcement Learning explores STAIRS-Former enhances multi-agent reinforcement learning by improving attention mechanisms for better coordination and long-horizon dependency capture.. Commercial viability score: 7/10 in Reinforcement Learning.