Main Content

rlPGAgent

Policy gradient reinforcement learning agent

Description

The policy gradient (PG) algorithm is a model-free, online, on-policy reinforcement learning method. A PG agent is a policy-based reinforcement learning agent that uses the REINFORCE algorithm to directly compute an optimal policy which maximizes the long-term reward. The action space can be either discrete or continuous.

For more information on PG agents and the REINFORCE algorithm, see Policy Gradient Agents. For more information on the different types of reinforcement learning agents, see Reinforcement Learning Agents.

Creation

Description

Create Agent from Observation and Action Specifications

example

agent = rlPGAgent(observationInfo,actionInfo) creates a policy gradient agent for an environment with the given observation and action specifications, using default initialization options. The actor and critic representations in the agent use default deep neural networks built from the observation specification observationInfo and the action specification actionInfo.

example

agent = rlPGAgent(observationInfo,actionInfo,initOpts) creates a policy gradient agent for an environment with the given observation and action specifications. The agent uses default networks in which each hidden fully connected layer has the number of units specified in the initOpts object. Policy gradient agents do not support recurrent neural networks. For more information on the initialization options, see rlAgentInitializationOptions.

Create Agent from Actor and Critic Representations

agent = rlPGAgent(actor) creates a PG agent with the specified actor network. By default, the UseBaseline property of the agent is false in this case.

agent = rlPGAgent(actor,critic) creates a PG agent with the specified actor and critic networks. By default, the UseBaseline option is true in this case.

Specify Agent Options

example

agent = rlPGAgent(___,agentOptions) creates a PG agent and sets the AgentOptions property to the agentOptions input argument. Use this syntax after any of the input arguments in the previous syntaxes.

Input Arguments

expand all

Observation specifications, specified as a reinforcement learning specification object or an array of specification objects defining properties such as dimensions, data type, and names of the observation signals.

You can extract observationInfo from an existing environment or agent using getObservationInfo. You can also construct the specifications manually using rlFiniteSetSpec or rlNumericSpec.

Action specifications, specified as a reinforcement learning specification object defining properties such as dimensions, data type, and names of the action signals.

For a discrete action space, you must specify actionInfo as an rlFiniteSetSpec object.

For a continuous action space, you must specify actionInfo as an rlNumericSpec object.

You can extract actionInfo from an existing environment or agent using getActionInfo. You can also construct the specification manually using rlFiniteSetSpec or rlNumericSpec.

Agent initialization options, specified as an rlAgentInitializationOptions object. Policy gradient agents do not support recurrent neural networks.

Actor network representation, specified as an rlStochasticActorRepresentation. For more information on creating actor representations, see Create Policy and Value Function Representations.

Critic network representation, specified as an rlValueRepresentation object. For more information on creating critic representations, see Create Policy and Value Function Representations.

Properties

expand all

Agent options, specified as an rlPGAgentOptions object.

Object Functions

trainTrain reinforcement learning agents within a specified environment
simSimulate trained reinforcement learning agents within specified environment
getActionObtain action from agent or actor representation given environment observations
getActorGet actor representation from reinforcement learning agent
setActorSet actor representation of reinforcement learning agent
getCriticGet critic representation from reinforcement learning agent
setCriticSet critic representation of reinforcement learning agent
generatePolicyFunctionCreate function that evaluates trained policy of reinforcement learning agent

Examples

collapse all

Create an environment with a discrete action space, and obtain its observation and action specifications. For this example, load the environment used in the example Create Agent Using Deep Network Designer and Train Using Image Observations. This environment has two observations: a 50-by-50 grayscale image and a scalar (the angular velocity of the pendulum). The action is a scalar with five possible elements (a torque of either -2, -1, 0, 1, or 2 Nm applied to the pole).

% load predefined environment
env = rlPredefinedEnv("SimplePendulumWithImage-Discrete");

% obtain observation and action specifications
obsInfo = getObservationInfo(env);
actInfo = getActionInfo(env);

The agent creation function initializes the actor and critic networks randomly. You can ensure reproducibility by fixing the seed of the random generator. To do so, uncomment the following line.

% rng(0)

Create a policy gradient agent from the environment observation and action specifications.

agent = rlPGAgent(obsInfo,actInfo);

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

getAction(agent,{rand(obsInfo(1).Dimension),rand(obsInfo(2).Dimension)})
ans = 1x1 cell array
    {[-2]}

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

Create an environment with a continuous action space and obtain its observation and action specifications. For this example, load the environment used in the example Train DDPG Agent to Swing Up and Balance Pendulum with Image Observation. This environment has two observations: a 50-by-50 grayscale image and a scalar (the angular velocity of the pendulum). The action is a scalar representing a torque ranging continuously from -2 to 2 Nm.

% load predefined environment
env = rlPredefinedEnv("SimplePendulumWithImage-Continuous");

% obtain observation and action specifications
obsInfo = getObservationInfo(env);
actInfo = getActionInfo(env);

Create an agent initialization option object, specifying that each hidden fully connected layer in the network must have 128 neurons (instead of the default number, 256). Policy gradient agents do not support recurrent networks, so setting the UseRNN option to true generates an error when the agent is created.

initOpts = rlAgentInitializationOptions('NumHiddenUnit',128);

The agent creation function initializes the actor and critic networks randomly. You can ensure reproducibility by fixing the seed of the random generator. To do so, uncomment the following line.

% rng(0)

Create a policy gradient agent from the environment observation and action specifications.

agent = rlPGAgent(obsInfo,actInfo,initOpts);

Reduce the critic learning rate to 1e-3.

critic = getCritic(agent);
critic.Options.LearnRate = 1e-3;
agent  = setCritic(agent,critic);

Extract the deep neural networks from both the agent actor and critic.

actorNet = getModel(getActor(agent));
criticNet = getModel(getCritic(agent));

Display the layers of the critic network, and verify that each hidden fully connected layer has 128 neurons

criticNet.Layers
ans = 
  11x1 Layer array with layers:

     1   'input_1'        Image Input       50x50x1 images
     2   'conv_1'         Convolution       64 3x3x1 convolutions with stride [1  1] and padding [0  0  0  0]
     3   'relu_input_1'   ReLU              ReLU
     4   'fc_1'           Fully Connected   128 fully connected layer
     5   'input_2'        Feature Input     1 features
     6   'fc_2'           Fully Connected   128 fully connected layer
     7   'concat'         Concatenation     Concatenation of 2 inputs along dimension 1
     8   'relu_body'      ReLU              ReLU
     9   'fc_body'        Fully Connected   128 fully connected layer
    10   'body_output'    ReLU              ReLU
    11   'output'         Fully Connected   1 fully connected layer

Plot actor and critic networks

plot(layerGraph(actorNet))

Figure contains an axes object. The axes object contains an object of type graphplot.

plot(layerGraph(criticNet))

Figure contains an axes object. The axes object contains an object of type graphplot.

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

getAction(agent,{rand(obsInfo(1).Dimension),rand(obsInfo(2).Dimension)})
ans = 1x1 cell array
    {[0.9228]}

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

Create an environment with a discrete action space, and obtain its observation and action specifications. For this example, load the environment used in the example Train PG Agent with Baseline to Control Double Integrator System. The observation from the environment is a vector containing the position and velocity of a mass. The action is a scalar representing a force, applied to the mass, having three possible values (-2, 0, or 2 Newton).

% load predefined environment
env = rlPredefinedEnv("DoubleIntegrator-Discrete");

% get observation and specification info
obsInfo = getObservationInfo(env);
actInfo = getActionInfo(env);

Create a critic representation to use as a baseline.

% create a network to be used as underlying critic approximator
baselineNetwork = [
    imageInputLayer([obsInfo.Dimension(1) 1 1], 'Normalization', 'none', 'Name', 'state')
    fullyConnectedLayer(8, 'Name', 'BaselineFC')
    reluLayer('Name', 'CriticRelu1')
    fullyConnectedLayer(1, 'Name', 'BaselineFC2', 'BiasLearnRateFactor', 0)];

% set some options for the critic
baselineOpts = rlRepresentationOptions('LearnRate',5e-3,'GradientThreshold',1);

% create the critic based on the network approximator
baseline = rlValueRepresentation(baselineNetwork,obsInfo,'Observation',{'state'},baselineOpts);

Create an actor representation.

% create a network to be used as underlying actor approximator
actorNetwork = [
    imageInputLayer([obsInfo.Dimension(1) 1 1], 'Normalization', 'none', 'Name', 'state')
    fullyConnectedLayer(numel(actInfo.Elements), 'Name', 'action', 'BiasLearnRateFactor', 0)];

% set some options for the actor
actorOpts = rlRepresentationOptions('LearnRate',5e-3,'GradientThreshold',1);

% create the actor based on the network approximator
actor = rlStochasticActorRepresentation(actorNetwork,obsInfo,actInfo,...
    'Observation',{'state'},actorOpts);

Specify agent options, and create a PG agent using the environment, actor, and critic.

agentOpts = rlPGAgentOptions(...
    'UseBaseline',true, ...
    'DiscountFactor', 0.99);
agent = rlPGAgent(actor,baseline,agentOpts)
agent = 
  rlPGAgent with properties:

    AgentOptions: [1x1 rl.option.rlPGAgentOptions]

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

getAction(agent,{rand(2,1)})
ans = 1x1 cell array
    {[-2]}

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

Create an environment with a continuous action space, and obtain its observation and action specifications. For this example, load the double integrator continuous action space environment used in the example Train DDPG Agent to Control Double Integrator System.

% load predefined environment
env = rlPredefinedEnv("DoubleIntegrator-Continuous");

% get observation specification info
obsInfo = getObservationInfo(env)
obsInfo = 
  rlNumericSpec with properties:

     LowerLimit: -Inf
     UpperLimit: Inf
           Name: "states"
    Description: "x, dx"
      Dimension: [2 1]
       DataType: "double"

% get action specification info
actInfo = getActionInfo(env)
actInfo = 
  rlNumericSpec with properties:

     LowerLimit: -Inf
     UpperLimit: Inf
           Name: "force"
    Description: [0x0 string]
      Dimension: [1 1]
       DataType: "double"

In this example, the action is a scalar input representing a force ranging from -2 to 2 Newton, so it is a good idea to set the upper and lower limit of the action signal accordingly. This must be done when the network representation for the actor has an nonlinear output layer than needs to be scaled accordingly to produce an output in the desired range.

% make sure action space upper and lower limits are finite
actInfo.LowerLimit=-2;
actInfo.UpperLimit=2;

Create a critic representation to use as a baseline. Policy gradient agents use a rlValueRepresentation for the baseline. For continuous observation spaces, you can use either a deep neural network or a custom basis representation. For this example, create a deep neural network as the underlying approximator.

% create a network to be used as underlying critic approximator
baselineNetwork = [
    imageInputLayer([obsInfo.Dimension 1], 'Normalization', 'none', 'Name', 'state')
    fullyConnectedLayer(8, 'Name', 'BaselineFC1')
    reluLayer('Name', 'Relu1')
    fullyConnectedLayer(1, 'Name', 'BaselineFC2', 'BiasLearnRateFactor', 0)];

% set some training options for the critic
baselineOpts = rlRepresentationOptions('LearnRate',5e-3,'GradientThreshold',1);

% create the critic based on the network approximator
baseline = rlValueRepresentation(baselineNetwork,obsInfo,'Observation',{'state'},baselineOpts);

Policy gradient agents use a rlStochasticActorRepresentation. For continuous action spaces stochastic actors, you can only use a neural network as underlying approximator.

The observation input (here called myobs) must accept a two-dimensional vector, as specified in obsInfo. The output (here called myact) must also be a two-dimensional vector (twice the number of dimensions specified in actInfo). The elements of the output vector represent, in sequence, all the means and all the standard deviations of every action (in this case there is only one mean value and one standard deviation).

The fact that standard deviations must be non-negative while mean values must fall within the output range means that the network must have two separate paths. The first path is for the mean values, and any output nonlinearity must be scaled so that it can produce outputs in the output range. The second path is for the variances, and you must use a softplus or relu layer to enforce non-negativity.

% input path layers (2 by 1 input, 1 by 1 output)
inPath = [ 
    imageInputLayer([obsInfo.Dimension 1], 'Normalization','none','Name','state')
    fullyConnectedLayer(10,'Name', 'ip_fc')   % 10 by 1 output
    reluLayer('Name', 'ip_relu')              % nonlinearity
    fullyConnectedLayer(1,'Name','ip_out') ]; % 1 by 1 output

% path layers for mean value (1 by 1 input and 1 by 1 output)
% using scalingLayer to scale the range
meanPath = [
    fullyConnectedLayer(15,'Name', 'mp_fc1') % 15 by 1 output
    reluLayer('Name', 'mp_relu')             % nonlinearity
    fullyConnectedLayer(1,'Name','mp_fc2');  % 1 by 1 output
    tanhLayer('Name','tanh');                % output range: (-1,1)
    scalingLayer('Name','mp_out','Scale',actInfo.UpperLimit) ]; % output range: (-2N,2N)

% path layers for standard deviation (1 by 1 input and output)
% using softplus layer to make it non negative
sdevPath = [
    fullyConnectedLayer(15,'Name', 'vp_fc1') % 15 by 1 output
    reluLayer('Name', 'vp_relu')             % nonlinearity
    fullyConnectedLayer(1,'Name','vp_fc2');  % 1 by 1 output
    softplusLayer('Name', 'vp_out') ];       % output range: (0,+Inf)

% conctatenate two inputs (along dimension #3) to form a single (2 by 1) output layer
outLayer = concatenationLayer(3,2,'Name','mean&sdev');

% add layers to layerGraph network object
actorNet = layerGraph(inPath);
actorNet = addLayers(actorNet,meanPath);
actorNet = addLayers(actorNet,sdevPath);
actorNet = addLayers(actorNet,outLayer);

% connect layers: the mean value path output MUST be connected to the FIRST input of the concatenation layer
actorNet = connectLayers(actorNet,'ip_out','mp_fc1/in');   % connect output of inPath to meanPath input
actorNet = connectLayers(actorNet,'ip_out','vp_fc1/in');   % connect output of inPath to variancePath input
actorNet = connectLayers(actorNet,'mp_out','mean&sdev/in1');% connect output of meanPath to mean&sdev input #1
actorNet = connectLayers(actorNet,'vp_out','mean&sdev/in2');% connect output of sdevPath to mean&sdev input #2

% plot network 
plot(actorNet)

Figure contains an axes object. The axes object contains an object of type graphplot.

Specify some options for the actor and create the stochastic actor representation using the deep neural network actorNet.

% set some options for the actor
actorOpts = rlRepresentationOptions('LearnRate',5e-3,'GradientThreshold',1);

% create the actor based on the network approximator
actor = rlStochasticActorRepresentation(actorNet,obsInfo,actInfo,...
    'Observation',{'state'},actorOpts);

Specify agent options, and create a PG agent using actor, baseline and agent options.

agentOpts = rlPGAgentOptions(...
    'UseBaseline',true, ...
    'DiscountFactor', 0.99);
agent = rlPGAgent(actor,baseline,agentOpts)
agent = 
  rlPGAgent with properties:

    AgentOptions: [1x1 rl.option.rlPGAgentOptions]

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

getAction(agent,{rand(2,1)})
ans = 1x1 cell array
    {[0.0347]}

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

For this example, load the environment used in the example Train PG Agent with Baseline to Control Double Integrator System. The observation from the environment is a vector containing the position and velocity of a mass. The action is a scalar representing a force, applied to the mass, having three possible values (-2, 0, or 2 Newton).

env = rlPredefinedEnv("DoubleIntegrator-Discrete");

Get observation and specification information.

obsInfo = getObservationInfo(env);
actInfo = getActionInfo(env);

Create a critic representation to use as a baseline. To create a recurrent neural network for the critic, use sequenceInputLayer as the input layer and include an lstmLayer as one of the other network layers.

baselineNetwork = [
    sequenceInputLayer(obsInfo.Dimension(1), 'Normalization', 'none', 'Name','myobs')
    fullyConnectedLayer(8, 'Name', 'BaselineFC')
    lstmLayer(8,'OutputMode','sequence','Name','lstm')
    reluLayer('Name', 'CriticRelu1')
    fullyConnectedLayer(1, 'Name', 'BaselineFC2', 'BiasLearnRateFactor', 0)];

Set some options for the critic.

baselineOpts = rlRepresentationOptions('LearnRate',5e-3,'GradientThreshold',1);

Create the critic based on the network approximator.

baseline = rlValueRepresentation(baselineNetwork,obsInfo,'Observation',{'myobs'},baselineOpts);

Create an actor representation. Since the critic has a recurrent network, the actor must have a recurrent network too.

Define a recurrent neural network for the actor.

actorNetwork = [
    sequenceInputLayer(obsInfo.Dimension(1), 'Normalization', 'none', 'Name', 'myobs')
    lstmLayer(8,'OutputMode','sequence','Name','lstm')
    fullyConnectedLayer(numel(actInfo.Elements), 'Name', 'action', 'BiasLearnRateFactor', 0)];

Set actor options and create the actor.

actorOpts = rlRepresentationOptions('LearnRate',5e-3,'GradientThreshold',1);
actor = rlStochasticActorRepresentation(actorNetwork,obsInfo,actInfo,...
    'Observation',{'myobs'},actorOpts);

Specify agent options, and create a PG agent using the environment, actor, and critic.

agentOpts = rlPGAgentOptions(...
    'UseBaseline',true, ...
    'DiscountFactor', 0.99);
agent = rlPGAgent(actor,baseline,agentOpts);

For PG agent with recurrent neural networks, the training sequence length is the whole episode.

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

getAction(agent,{obsInfo.Dimension})
ans = 1×1 cell array
    {[0]}

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

Tips

  • For continuous action spaces, the rlPGAgent agent does not enforce the constraints set by the action specification, so you must enforce action space constraints within the environment.

Introduced in R2019a