Assume that you have an existing trained reinforcement learning agent. For this example, load the trained agent from Train DDPG Agent to Control Double Integrator System.

load('DoubleIntegDDPG.mat','agent') 

Obtain the actor representation from the agent.

actor = getActor(agent);

Obtain the learnable parameters from the actor.

params = getLearnableParameters(actor);

Modify the parameter values. For this example, simply multiply all of the parameters by 2.

modifiedParams = cellfun(@(x) x*2,params,'UniformOutput',false);

Set the parameter values of the actor to the new modified values.

actor = setLearnableParameters(actor,modifiedParams);

Set the actor in the agent to the new modified actor.

agent = setActor(agent,actor);

Assume that you have an existing reinforcement learning agent, agent. For this example, load the trained agent from Train DDPG Agent to Control Double Integrator System:

load('DoubleIntegDDPG.mat','agent')

Further, assume that this agent has an actor representation that contains the following shallow neural network structure:

oldActorNetwork = [
        imageInputLayer([2 1 1],'Normalization','none','Name','state')
        fullyConnectedLayer(1,'Name','action')];

Create the new network with the additional fully connected hidden layer:

newActorNetwork = [
        imageInputLayer([2 1 1],'Normalization','none','Name','state')
        fullyConnectedLayer(3,'Name','hidden');
        fullyConnectedLayer(1,'Name','action')];

Create the corresponding actor representation:

actor = rlDeterministicActorRepresentation(newActorNetwork,...
    getObservationInfo(agent),getActionInfo(agent),...
    'Observation',{'state'},...
    'Action',{'action'})

actor = 
  rlDeterministicActorRepresentation with properties:

         ActionInfo: [1x1 rl.util.rlNumericSpec]
    ObservationInfo: [1x1 rl.util.rlNumericSpec]
            Options: [1x1 rl.option.rlRepresentationOptions]

Set the actor representation of the agent to the new augmented actor:

agent = setActor(agent,actor);

To check your agent, use getAction to return the action from a random observation.

getAction(agent,{rand(2,1)})

ans = single
    1.4134

You can now test and train the agent against the environment.