How to setup a multi-agent DDPG

Hi,

I am trying to simulate a number of agents that collaboratively doing mapping. I designed the actor critic networks, but I have a problem that how I can write a code for gridworld file inside my simulink. Is there any related example?

Answer 1

[Alan]

Hi Ali,

The following documentation page shows an example with multiple agents that perform a collaborative task.: https://www.mathworks.com/help/releases/R2022a/reinforcement-learning/ug/train-2-agents-to-collaborate.html

Make sure that you declare the environment by calling rlSimulinkEnv with the multiple RL block paths and their associated observation and action information as shown below:

blks = ["model_name/Agent A (Red)", "model_name/Agent B (Green)", "model_name/Agent C (Blue)", "model_name/Agent D (Black)", "model_name/Agent E (Yellow)"];
obsInfos = {oinfo1, oinfo2, oinfo3, oinfo4, oinfo5};
actInfos = {ainfo1, ainfo2, ainfo3, ainfo4, ainfo5};
env = rlSimulinkEnv(mdl, blks, obsInfos, actInfos);

For more detailed information on creating custom Simulink Environments for Reinforcement Learning, you can refer to the following documentation page: https://www.mathworks.com/help/releases/R2022a/reinforcement-learning/ug/create-simulink-environments-for-reinforcement-learning.html

- Alan