Research map / Multi-agent systems and games / Human-AI and goal-conditioned
ELF: An Extensive, Lightweight and Flexible Research Platform for Real-time Strategy Games
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 |
|---|---|---|---|
| Add Leaky ReLU and BatchNorm to the A3C network (ablation) | A3C on Mini-RTS · Mini-RTS vs SIMPLE and HIT_N_RUN | win rate (%): ReLU: 54.7 ($\pm$ 4.2), 57.0 ($\pm$ 6.8); Leaky ReLU + BN: 68.4 ($\pm$ 4.3), 63.6 ($\pm$ 7.9) | Helped |
| Add Leaky ReLU to BatchNorm (ablation) | A3C on Mini-RTS · Mini-RTS vs SIMPLE and HIT_N_RUN | win rate (%): BN: 64.4 ($\pm$ 7.4 ), 57.5 ($\pm$ 6.8); Leaky ReLU + BN: 68.4 ($\pm$ 4.3), 63.6 ($\pm$ 7.9) | Mixed |
| Add curriculum training (ablation) | A3C · Mini-RTS vs SIMPLE and HIT_N_RUN; Capture the Flag | win rate (%): 66.0($\pm$2.4) -> 68.4 ($\pm$4.3); 54.4($\pm$15.9) -> 63.6 ($\pm$7.9); 54.2($\pm$20.0) -> 59.9 ($\pm$7.4) | Helped |
| Add curriculum when training against a mixture of opponents (ablation) | A3C on Mini-RTS · Mini-RTS vs combined SIMPLE and HIT_N_RUN | win rate (%): 47.7($\pm$11.0) -> 53.2($\pm$8.5) | Helped |
| Train against a mixture of opponent strategies rather than only one | A3C on Mini-RTS · Mini-RTS vs combined SIMPLE and HIT_N_RUN | win rate (%): 47.5($\pm$5.1), 49.1($\pm$10.5) -> 53.2($\pm$8.5) | Helped |
| Increase opponent frame skip from 10 to 50 | A3C on Mini-RTS · Mini-RTS vs SIMPLE and HIT_N_RUN | win rate (%): 52.8($\pm$2.4) -> 68.4($\pm$4.3); 51.1($\pm$5.0) -> 63.6($\pm$7.9) | Helped |
| Use longer A3C history length (ablation) | A3C on Mini-RTS · Mini-RTS | win rate | Helped |
| Train a full-game agent with A3C rather than a random policy | A3C · Capture the Flag and Tower Defense | win rate (%): 0.7 ($\pm$ 0.9) -> 59.9 ($\pm$ 7.4); 36.3 ($\pm$ 0.3) -> 91.0 ($\pm$ 7.6) | Helped |
| Train against an opponent using a different strategy rather than the evaluated opponent | A3C on Mini-RTS · Mini-RTS vs the other built-in AI | win rate (%): 34.6($\pm$13.1) vs SIMPLE; 26.6($\pm$7.6) vs HIT_N_RUN | Hurt |
| Use MCTS planning with full game-state access | MCTS on Mini-RTS · Mini-RTS vs SIMPLE and HIT_N_RUN | win rate (%): Random: 24.2($\pm$3.9), 25.9($\pm$0.6); MCTS: 73.2($\pm$0.6), 62.7($\pm$2.0) | Helped |
Closest papers by meaning
- TowerMind: A Tower Defence Game Learning Environment and Benchmark for LLM as Agents (AAAI 2026 · similarity 0.85)
- JaxMARL: Multi-Agent RL Environments and Algorithms in JAX (NeurIPS 2024 · similarity 0.85)
- Craftax: A Lightning-Fast Benchmark for Open-Ended Reinforcement Learning (ICML 2024 · similarity 0.84)
- FightLadder: A Benchmark for Competitive Multi-Agent Reinforcement Learning (ICML 2024 · similarity 0.84)
- Strategist: Self-improvement of LLM Decision Making via Bi-Level Tree Search (ICLR 2025 · similarity 0.84)
- CPMöbius: Iterative Coach–Player Reasoning for Data-Free Reinforcement Learning (ICML 2026 · 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.