OpenAI Five — Dota 2 milestone.
hardAnswer
- Five-agent RL (one per hero) trained via self-play on massive scale (128k CPU cores, 256 GPUs).
- PPO with LSTM policy, shared parameters across heroes with role embedding.
- Reward shaping via team + individual signals.
- Coordination is emergent — no explicit coordination mechanism.
- Defeated world-champion team OG in 2019.
- Foundation of large-scale team-game MARL.
Check yourself — multiple choice
- Random
- 5-agent RL (one per hero) via PPO + LSTM + role embed + massive-scale self-play (128k CPU / 256 GPU); reward shaping team + individual; defeated OG 2019 — MARL milestone
- Same as AlphaStar
- Not real
OpenAI Five: PPO + LSTM + massive self-play + team + individual rewards → OG 2019.
#multi-agent#deep-rl
Practise Reinforcement Learning
214 interview questions in this topic.