Research map / Reinforcement learning / Visual and representation RL
On Reinforcement Learning for Full-Length Game of StarCraft
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 |
|---|---|---|---|
| combine combat network targeting with fixed-position combat-rule attacks (ablation) | Hierarchical StarCraft II agent · SC2LE simple64, difficulty levels 1-9 | winning rate | Helped |
| train first on lower difficulty and transfer the model to higher difficulty (ablation) | Combat Network agent · SC2LE difficulty level-7 | winning rate | Helped |
| initialize combat network training with controller and base-network parameters trained using combat rule (ablation) | Combat Network agent · SC2LE difficulty level-2 | learning performance | Helped |
| update controller and sub-policies alternately rather than simultaneously (ablation) | Hierarchical agent · SC2LE difficulty level-7 | training stability | Helped |
| use hierarchical controller and sub-policies rather than a single policy while retaining macro-actions (ablation) | StarCraft II agent · SC2LE difficulty level-7 | final performance | Helped |
| retain expert-derived macro-actions while removing the controller and sub-policies (ablation) | StarCraft II agent · SC2LE low difficulty levels | learning performance | No clear effect |
| use Win/Loss reward instead of handcrafted rewards (ablation) | Hierarchical Combat Network agent · SC2LE difficulty level-7 | performance | Hurt |
| use Blizzard score as reward instead of win-oriented handcrafted rewards (ablation) | Combat Network agent · SC2LE difficulty level-2 | winning rate | Hurt |
| increase episodes collected per iteration (ablation) | Combat Rule agent · SC2LE difficulty level-2 | training stability | Helped |
| use combat rule instead of combat network (ablation) | StarCraft II agent · SC2LE difficulty levels 1-9 | winning rate | Mixed |
Closest papers by meaning
- SCC: an efficient deep reinforcement learning agent mastering the game of StarCraft II (ICML 2021 · similarity 0.90)
- Episodic Exploration for Deep Deterministic Policies for StarCraft Micromanagement (ICLR 2017 · similarity 0.90)
- Grid-Wise Control for Multi-Agent Reinforcement Learning in Video Game AI (ICML 2019 · similarity 0.87)
- LLM-PySC2: Starcraft II learning environment for Large Language Models (NeurIPS 2025 · similarity 0.85)
- Accelerate Multi-Agent Reinforcement Learning in Zero-Sum Games with Subgame Curriculum Learning (AAAI 2024 · similarity 0.85)
- Mimicking To Dominate: Imitation Learning Strategies for Success in Multiagent Games (NeurIPS 2024 · similarity 0.85)
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.