Research map / Multi-agent systems and games / Games and equilibria
SCC: an efficient deep reinforcement learning agent mastering the game of StarCraft II
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 |
|---|---|---|---|
| Train with 4,638 replays instead of the full 105,034-replay dataset (ablation) | SCC supervised learning · Built-in Elite bot evaluation | win rate | No clear effect |
| Use higher-MMR replays (MMR5300) instead of the MMR4300 medium dataset (ablation) | SCC supervised learning · Built-in Elite bot evaluation | win rate: around 85%, about 10% higher | Helped |
| Double batch size from 3,850 to 7,700 (ablation) | SCC supervised learning on the large dataset · Built-in Elite bot evaluation | win rate | Helped |
| Fine-tune on 3,458 high-quality winning replays after full-dataset training (ablation) | SCC supervised learning · Built-in Elite bot evaluation | win rate: 97% | Helped |
| Reduce minimap size from 128x128 to 64x64 (ablation) | SCC policy network · Supervised learning | performance | No clear effect |
| Replace average pooling with trainable attention-based pooling (ablation) | SCC policy network · Supervised learning | performance | Helped |
| Use agent branching: initialize an exploiter from the current main agent and guide it with strategy-statistic reward (ablation) | SCC league training · StarCraft II robustness and exploiter training | training efficiency | Helped |
| Train multiple main agents rather than one (ablation) | SCC league training · StarCraft II strategy robustness | strategy diversity | Helped |
| Apply reinforcement learning after supervised initialization | SCC · Built-in bots across races and maps | win rate: 0.97 -> 1.0 | Helped |
Closest papers by meaning
- On Reinforcement Learning for Full-Length Game of StarCraft (AAAI 2019 · similarity 0.90)
- Episodic Exploration for Deep Deterministic Policies for StarCraft Micromanagement (ICLR 2017 · similarity 0.89)
- Mimicking To Dominate: Imitation Learning Strategies for Success in Multiagent Games (NeurIPS 2024 · similarity 0.84)
- LLM-PySC2: Starcraft II learning environment for Large Language Models (NeurIPS 2025 · similarity 0.84)
- FightLadder: A Benchmark for Competitive Multi-Agent Reinforcement Learning (ICML 2024 · similarity 0.84)
- JaxMARL: Multi-Agent RL Environments and Algorithms in JAX (NeurIPS 2024 · similarity 0.83)
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.