A new transformer-based multi-agent world model with teammate prediction and prioritized replay achieves near-optimal performance on cooperative benchmarks in as few as 50,000 environment steps.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Transformer World Model for Sample Efficient Multi-Agent Reinforcement Learning
A new transformer-based multi-agent world model with teammate prediction and prioritized replay achieves near-optimal performance on cooperative benchmarks in as few as 50,000 environment steps.