Research map / Multi-agent systems and games / Games and equilibria
Towards Playing Full MOBA Games with Deep Reinforcement Learning
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 curriculum self-play learning (fixed-lineup teachers, distillation, merged training) instead of random lineups (ablation) | MOBA AI training, 40 heroes · Honor of Kings training | training time / convergence: > 480 h -> 336 h | Helped |
| use CSPL instead of random lineups (ablation) | MOBA AI training, 20 heroes · Honor of Kings training | training time (h): 192 h -> 144 h | Helped |
| add curriculum self-play learning (ablation) | final 40-hero AI · Honor of Kings Elo ablation | Elo | Helped |
| add multi-head value estimation (ablation) | value function · Honor of Kings Elo ablation | Elo | Helped |
| feed invisible opponent information to value function (ablation) | value function · Honor of Kings Elo ablation | Elo | Helped |
| use Dual-clip PPO instead of standard PPO (ablation) | policy updates · Honor of Kings Elo ablation | Elo | Helped |
| use MCTS-based drafting instead of random hero picking (ablation) | drafting agent · Honor of Kings drafting, 1000 matches | mean win rate: 0.86 | Helped |
| use MCTS-based drafting instead of highest win-rate picking (ablation) | drafting agent · Honor of Kings drafting, 1000 matches | mean win rate: 0.68 | Helped |
| policy distillation of fixed-lineup teachers into one student (ablation) | CSPL phase 2 · Honor of Kings distillation | Elo | Mixed |
Closest papers by meaning
- Mastering Complex Control in MOBA Games with Deep Reinforcement Learning (AAAI 2020 · similarity 0.93)
- Learning Diverse Policies in MOBA Games via Macro-Goals (NeurIPS 2021 · similarity 0.91)
- Hierarchical Macro Strategy Model for MOBA Game AI (AAAI 2019 · similarity 0.89)
- Large Scale Learning of Agent Rationality in Two-Player Zero-Sum Games (AAAI 2019 · similarity 0.87)
- Finding Friend and Foe in Multi-Agent Games (NeurIPS 2019 · similarity 0.87)
- FightLadder: A Benchmark for Competitive Multi-Agent Reinforcement Learning (ICML 2024 · similarity 0.86)
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.