Research map / Multi-agent systems and games / Cooperative multi-agent RL
Learning from Good Trajectories in Offline Multi-Agent Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use SIT to decompose joint data into agent-wise trajectories and share high-quality trajectories | Offline MARL · StarCraft II 3s_vs_5z, low-quality dataset | episode return: 5.2$\pm$0.1 -> 11.1$\bm{\pm}$0.8 | Helped |
| Use SIT to decompose joint data into agent-wise trajectories and share high-quality trajectories | Offline MARL · MPE CN_3ls3l, low-quality dataset | episode return: -117.8$\pm$11.5 -> -92.4$\bm{\pm}$7.2 | Helped |
| Use SIT to decompose joint data into agent-wise trajectories and share high-quality trajectories | Offline MARL · MAmujoco HalfCheetah_2l, low-quality dataset | episode return: -109.3$\pm$3.1 -> -0.3$\bm{\pm}$0.1 | Helped |
| Remove reward decomposition and train the critic directly on global reward (ablation) | SIT · StarCraft II 2s3z, low-quality dataset | performance | Hurt |
| Use equal weights for all estimated individual rewards instead of learned attention weights (ablation) | SIT · StarCraft II 2s3z, low-quality dataset | performance | Hurt |
| Remove the ensemble mechanism from reward decomposition (ablation) | SIT · StarCraft II 2s3z, low-quality dataset | performance | Hurt |
| Remove GAT aggregation from the critic (ablation) | SIT · StarCraft II 2s3z, low-quality dataset | performance | Hurt |
| Use priority sampling to favor high-quality individual trajectories (ablation) | SIT · StarCraft II 2s3z, extreme low-quality dataset | performance | Helped |
| Replace attention-based reward decomposition with a monotonic nonlinear constraint decomposition (ablation) | SIT · StarCraft II 2s3z, low-quality dataset | performance | Hurt |
| Use SIT on agent-wise balanced data rather than imbalanced data (ablation) | SIT · StarCraft II 2s3z, balanced datasets | performance | No clear effect |
Closest papers by meaning
- Offline Multi-Agent Reinforcement Learning via In-Sample Sequential Policy Optimization (AAAI 2025 · similarity 0.92)
- Dispelling the Mirage of Progress in Offline MARL through Standardised Baselines and Evaluation (NeurIPS 2024 · similarity 0.91)
- ComaDICE: Offline Cooperative Multi-Agent Reinforcement Learning with Stationary Distribution Shift Regularization (ICLR 2025 · similarity 0.91)
- Cooperative Policy Agreement: Learning Diverse Policy for Offline MARL (AAAI 2025 · similarity 0.90)
- Learning Generalizable Skills from Offline Multi-Task Data for Multi-Agent Cooperation (ICLR 2025 · similarity 0.89)
- Exploiting Structure in Offline Multi-Agent RL: The Benefits of Low Interaction Rank (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.