How to create a neural network for Multiple Agent with discrete and continuous action?
2 views (last 30 days)
Show older comments
Janani Sunil
on 26 Apr 2021
Answered: Emmanouil Tzorakoleftherakis
on 26 Apr 2021
Hi All,
I am trying to create a RL model with 2 agents in my environment.
Both the observations are continuous, but Agent 1 Action is discrete and Agent 2 actions are continuous. How do I specify them while building the actor network?
%Create Action Specifications
numActions = 3;
numActions2 = 1;
actionSizes = numActions+ numActions2
numActionCombinations = 8;
S0 = [0 0 0];
S1 = [0 0 1];
S2 = [0 1 1];
S3 = [0 1 0];
S4 = [1 1 0];
S5 = [1 0 1];
S6 = [1 0 0];
S7 = [1 1 1];
actionInfo = rlFiniteSetSpec({S0,S1,S2,S3,S4,S5,S6,S7});
actionInfo2 = rlNumericSpec([numActions2 1],'LowerLimit',0.05,'UpperLimit',30);
actionInfo.Name = 'Pulse';
actionInfo2.Name = 'cRef';
net = [ featureInputLayer(obsSizes,'Normalization','none','Name','state')
fullyConnectedLayer(actionSizes,'Name','fc')
softmaxLayer('Name','actionProb') ];
actor = rlStochasticActorRepresentation(net,obsInfo,actInfo,'Observation','state');
0 Comments
Accepted Answer
Emmanouil Tzorakoleftherakis
on 26 Apr 2021
If you want to specify the neural network structures yourself, there is nothing specific you need to do - simply create two actors and two critics, one for each action space and you are all set.
There is also the option to use the default agent feature where the neural nets are created automatically for you by only providing the observation and action space. See an example here.
0 Comments
More Answers (0)
See Also
Categories
Find more on Environments in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!