openai / multiagent-particle-envs

Code for a multi-agent particle environment used in the paper "Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments"
https://arxiv.org/pdf/1706.02275.pdf
MIT License
2.27k stars 786 forks source link

Questions about simple_spread agent actions #93

Open zhouweiqing-star opened 2 years ago

zhouweiqing-star commented 2 years ago

Hello ,dear author, I saw someone on github reproduced maddpg in the mpe environment, the joint action generated each time in him is [array([1.3097883e-02, 9.7529411e-01, 4.0262565e-03, 6.8232100e -03, 7.5863604e-04], dtype = float32), array ([0.11447875, 0.01806207, 0.00564417, 0.04932265, 0.8124923], dtype = float32), array ([0.23820595, 0.5213099, 0.12905039, 0.07524598, 0.03618779], dtype = float32 )], why does the action of an agent correspond to array([1.3097883e-02, 9.7529411e-01, 4.0262565e-03, 6.8232100e-03, 7.5863604e-04], which contains 5 values, what do they mean? ?