摘自:
算法中心-人机对抗智能 
1.对抗空间表示
1.1特征表示
基于卷积神经网络的地图特征提取算法
基于全连接的算子特征提取算法
Transformer
LSTM(Long Short-Term Memory)
1.2奖励函数
RND(random network distillation)
Hindsight
2.态势评估推理
意图判定:intentMARL
对手建模:DRON(Deep Reinforcement Opponent Network)
未来预测:DFP
威胁评估\透视分析
3.策略生成优化.
3.1零和博弈
Minimax Q-Learning
MCTS(Monte Carlo Tree Search)
RTNS博弈算法[1] Korf, Richard E., and David Maxwell Chickering. "Best-first minimax search." Artificial intelligence 84.1-2 (1996): 299-337.
3.2非零和博弈
FFQ(Friend-or-Foe Q-Learning)
Nash Q-Learning
3.3完美信息博弈
MCTS(Monte Carlo Tree Search
Double DQN(Double Deep Q-Networks
DDPG(Deep Deterministic Policy Gradient
AC(Actor-Critic
TRPO(Trust Region Policy Optimization
IS-MCTS(Information Set Monte Carlo Tree Search
AC(Actor-Critic
3.4非完美信息博弈
NFSP(Neural Fictitious Self-Play
CFR-BR(CFR against a best responder
临时价值评估（EVA）[1] Hansen, Steven, et al. "Fast deep reinforcement learning using online adjustments from the past." Advances in Neural Information Processing Systems. 2018.
External sampling MCCFR(External sampling Monte Carlo CFR
Best Response
CFR：反事实后悔最小化算法（Counterfactual Regret Minimization）
Outcome sampling Monte Carlo CFR 反事实后悔最小化（CFR）
Deep CFR：Deep CFR
RPG：Regret Policy Gradient
TRPO(Trust Region Policy Optimization
3.5人物博弈
PPO(Proximal Policy Optimization
DQN
Double DQN
Dueling DQN
SARSA(State-Action-Reward-State-Action
Sarsa (Lambda)
DDPG(Deep Deterministic Policy Gradient
AC
TRPO：Trust Region Policy Optimization
PER-DQN：Prioritized Experience Replay
Local GAC：Local Generative Actor-Critic
3.6两人博弈
Minimax Q-Learning
DQN
Neural Fictitious Self-Play
MCTS
DDPG(Deep Deterministic Policy Gradient
AC
TRPO(Trust Region Policy Optimization
Dueling DQN
SARSA：State-Action-Reward-State-Action
Sarsa (Lambda)
PER-DQN
Local GAC
RTNS
3.7多人博弈
MADDPG(Multi-agent actor-critic for mixed cooperative-competitive environments
MFMARL：Mean Field Multi-Agent Reinforcement Learning
NFSP：Neural Fictitious Self-Play
DRON(Deep Reinforcement Opponent Network
Nash Q：Nash Q-Learning
CFR：反事实后悔最小化算法（Counterfactual Regret Minimization）
RPG：Regret Policy Gradient
4.行动协同控制
4.1路径搜索
A*（中文通常读作A星，英文读作A-Star）
D*(Dynamic A*)
4.2多智能体协同
VDN：Value-Decomposition Networks
CoPPO：[1] Chao Yu, Akash Velu, Eugene Vinitsky, Yu Wang, A. Bayen, and Yi Wu. The surprising effectiveness 363 of mappo in cooperative, multi-agent games. ArXiv, abs/2103.01955, 2021.
MARL：[1] Tabish Rashid, Mikayel Samvelyan, Christian Schroeder de Witt, Gregory Farquhar, Jakob Foerster, Shimon Whiteson: “QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning”, 2018; arXiv:1803.11485.
