Research map / Multi-agent systems and games / Cooperative multi-agent RL
Hierarchical Multi-Agent Skill Discovery
Mingyu Yang, Yaodong Yang, Zhenbo Lu, Wengang Zhou, Houqiang Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 HMASD instead of EITI exploration bonus | MARL baseline · Overcooked cramped_room | final episode reward: 0.122 -> 236.0 | Improved |
| use HMASD instead of EDTI exploration bonus | MARL baseline · Overcooked coordination_ring | final episode reward: 0.0 -> 80.0 | Improved |
| use HMASD instead of EITI | MARL baseline · Overcooked asymmetric_advantages | final episode reward: 0.008 -> 295.1 | Improved |
| learn both team and individual skills (vs team skill only) (ablation) | HMASD · SMAC 0-1 reward and Overcooked | performance | Improved |
| add extrinsic team reward to low-level reward (ablation) | HMASD · SMAC 0-1 reward and Overcooked | performance | Improved |
| add intrinsic diversity rewards to low-level reward (ablation) | HMASD · SMAC 0-1 reward and Overcooked | performance | Improved |
| add high-level skill coordinator (vs random skill assignment per episode) (ablation) | HMASD · SMAC 0-1 reward and Overcooked | performance | Improved |
| vary skill interval k (ablation) | HMASD · ablation Fig. 10 | performance | Mixed |
| vary number of team and individual skills (ablation) | HMASD · ablation Fig. 10 | performance | Mixed |
| use HMASD instead of MAPPO and MAT | MAPPO, MAT · SMAC with 0-1 reward | win rate | Improved |
| use HMASD instead of MASER | MASER · SMAC with 0-1 reward | win rate | Improved |
| increase state-observation space size | HMASD · SMAC | useful skills fraction: 12 of 50 skills useful, 24% | Hurt |
Closest papers by meaning
- Learning Generalizable Skills from Offline Multi-Task Data for Multi-Agent Cooperation (ICLR 2025 · similarity 0.91)
- Disentangled Unsupervised Skill Discovery for Efficient Hierarchical Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- SkiLD: Unsupervised Skill Discovery Guided by Factor Interactions (NeurIPS 2024 · similarity 0.87)
- SkillMaster: Toward Autonomous Skill Mastery in LLM Agents (NeurIPS 2026 · similarity 0.87)
- Constrained Ensemble Exploration for Unsupervised Skill Discovery (ICML 2024 · similarity 0.87)
- Efficient Skill Discovery via Regret-Aware Optimization (ICML 2025 · similarity 0.87)
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.
