EasyDeepLearn

MAPPO — Multi-Agent PPO.

medium

Answer

  • Yu et al. 2022.
  • Simply PPO with a centralized value function using joint observations, shared across cooperative agents.
  • Surprisingly strong: matches or beats specialized MARL algorithms (QMIX, MADDPG) on many benchmarks (SMAC, MPE) with careful implementation.
  • Foundation of 'PPO is enough' school of MARL.
  • Simpler than QMIX/MADDPG; recommended default.
Check yourself — multiple choice
  • Random
  • PPO with centralized value function using joint obs; matches/beats QMIX/MADDPG on SMAC/MPE benchmarks with careful implementation; recommended cooperative MARL default
  • Same as PPO
  • Not real

MAPPO: PPO + centralized critic on joint obs; strong MARL default.

#multi-agent#actor-critic

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions