Research map / Multi-agent systems and games / Cooperative multi-agent RL
Google Research Football: A Novel Reinforcement Learning Environment
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 |
|---|---|---|---|
| Use Checkpoint rather than Scoring reward | IMPALA · Football Benchmarks, especially medium and hard | average goal difference: -0.36 ± 0.11 -> 5.54 ± 0.90; -0.47 ± 0.48 -> 3.15 ± 0.37 | Helped |
| Use Checkpoint rather than Scoring reward | Ape-X DQN · Football Benchmarks | average goal difference: 2.01 ± 0.27 -> 2.18 ± 0.25; 0.27 ± 0.56 -> 1.20 ± 0.40 | No clear effect |
| Train against Agent A instead of the built-in AI | Agent B · 11 vs 11 medium, B vs built-in AI | average goal difference: 4.25 ± 1.72 -> -0.27 ± 0.33 | Hurt |
| Control more players (1 to 3) | IMPALA · 3 versus 1 with Keeper, 50M steps | score: 0.68 ± 0.03 -> 0.86 ± 0.08 | Helped |
| Control more players (1 to 3) | IMPALA · 3 versus 1 with Keeper, 5M steps | score: 0.38 ± 0.23 -> 0.26 ± 0.11 | Mixed |
| Use grayscale pixels rather than Floats | IMPALA · Football Benchmarks Easy, Checkpoint, 100M steps | average goal advantage: 2.42 ± 0.46 -> -0.82 ± 0.26 | Hurt |
| Use grayscale pixels rather than Floats | IMPALA · Football Benchmarks Easy, Checkpoint, 500M steps | average goal advantage: 5.73 ± 0.28 -> 7.18 ± 0.85 | Helped |
| Stack the SMM representation | IMPALA · Football Benchmarks Easy, Checkpoint, 100M steps | average goal advantage: 5.94 ± 0.86 -> 7.62 ± 0.67 | Helped |
| Stack the SMM representation | IMPALA · Football Benchmarks Easy, Checkpoint, 500M steps | average goal advantage: 9.75 ± 4.15 -> 12.89 ± 0.51 | Helped |
| Increase training from 100M to 500M steps with grayscale pixels | IMPALA · Football Benchmarks Easy, Checkpoint | average goal advantage: -0.82 ± 0.26 -> 7.18 ± 0.85 | Helped |
Closest papers by meaning
- The CoachAI Badminton Environment: A Novel Reinforcement Learning Environment with Realistic Opponents (Student Abstract) (AAAI 2024 · similarity 0.86)
- The CoachAI Badminton Environment: Bridging the Gap between a Reinforcement Learning Environment and Real-World Badminton Games (AAAI 2024 · similarity 0.86)
- Policy Learning from Tutorial Books via Understanding, Rehearsing and Introspecting (NeurIPS 2024 · similarity 0.85)
- FightLadder: A Benchmark for Competitive Multi-Agent Reinforcement Learning (ICML 2024 · similarity 0.85)
- Foundations of Multi-Agent Learning in Dynamic Environments: Where Reinforcement Learning Meets Strategic Decision-Making (AAAI 2025 · similarity 0.84)
- Towards Building Human-like Smart Agents in Modern 3D Video Games (Student Abstract) (AAAI 2025 · 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.