Research map / Multi-agent systems and games / Human-AI and goal-conditioned
A Robust and Opponent-Aware League Training Method for StarCraft II
Ruozi Huang, Xipeng Wu, Hongsheng Yu, Zhong Fan, Haobo Fu, QIANG FU, Yang Wei
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 |
|---|---|---|---|
| add opponent modeling embedding and scouting reward to AlphaStar (ablation) | AlphaStar MA · VC Stalker Play opponent | win rate: 22% -> 52% | Improved |
| add opponent modeling to AlphaStar (ablation) | AlphaStar MA · Void Ray push opponent | win rate: 54% -> 60% | Improved |
| add opponent modeling to AlphaStar (ablation) | AlphaStar MA · Proxy VD opponent | win rate: 50% -> 56% | Improved |
| add opponent modeling to AlphaStar (ablation) | AlphaStar MA · Disruptor push opponent | win rate: 48% -> 58% | Improved |
| scouting reward from opponent prediction (ablation) | AlphaStar + opponent modeling · time to discover opponent buildings | discovery time | Improved |
| replace AlphaStar with full ROA-Star (goal-conditioned exploiters + opponent modeling) | AlphaStar · 10-day league | worst-case win rate: 1% -> 44% | Improved |
| full ROA-Star vs AlphaStar average win rate | AlphaStar · 10-day league | average win rate: 70.2% -> 84.1% | Improved |
| full ROA-Star vs AlphaStar RPP | AlphaStar · 10-day league | RPP score: 0.2157 -> 0.7843 | Improved |
| train exploiters on random z instead of unconditioned (ablation) | AlphaStar · 5-day MA robustness | worst-case win rate / Elo | Improved |
| use goal-conditioned EIE+ERE instead of random-z exploiters (ablation) | AlphaStar + random-z exploiters · 5-day MA robustness | worst-case win rate / Elo | Improved |
| add EIE on top of ERE-only exploiters (ablation) | AlphaStar + ERE · 5-day MA Elo | Elo | Improved |
| use curricular reward and strategy-guided loss in ERE instead of edit-distance reward (ablation) | Explorative Exploiters · target build order execution | last-entity execution precision | Improved |
Closest papers by meaning
- Designing Skill-Compatible AI: Methodologies and Frameworks in Chess (ICLR 2024 · similarity 0.86)
- Maia-2: A Unified Model for Human-AI Alignment in Chess (NeurIPS 2024 · similarity 0.85)
- Human-Aligned Chess With a Bit of Search (ICLR 2025 · similarity 0.85)
- Mimicking To Dominate: Imitation Learning Strategies for Success in Multiagent Games (NeurIPS 2024 · similarity 0.84)
- Enhancing Human Experience in Human-Agent Collaboration: A Human-Centered Modeling Approach Based on Positive Human Gain (ICLR 2024 · similarity 0.84)
- The CoachAI Badminton Environment: Bridging the Gap between a Reinforcement Learning Environment and Real-World Badminton Games (AAAI 2024 · similarity 0.84)
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.
