main content

create environment object from a simulink environment model that does not contain an agent block -凯发k8网页登录

create environment object from a simulink environment model that does not contain an agent block

since r2019a

syntax

env = createintegratedenv(refmodel,newmodel)
[env,agentblock,obsinfo,actinfo] = createintegratedenv(___)
[___] = createintegratedenv(___,name=value)

description

given a simulink® environment model that does not include your agent block, the createintegratedenv function generates a new closed-loop simulink model that contains an agent block and references your original environment model from its environment block. the function also returns an environment object that you can use for training and simulation. the environment object acts as an interface so that when you call sim or train, these functions in turn call the created (and compiled) simulink model to generate experiences for the agents.

to create an environment object from a simulink model that already includes an agent block, use rlsimulinkenv instead. for more information on simulink reinforcement learning environments, see create custom simulink environments.

env = createintegratedenv(refmodel,newmodel) creates a simulink model with the name specified by newmodel and returns a reinforcement learning environment object, env, for this model. the new model contains an rl agent block and references refmodel within its environment block. for more information on model referencing, see (simulink).

example

[env,agentblock,obsinfo,actinfo] = createintegratedenv(___) returns the block path to the rl agent block in the new model and the observation and action data specifications for the reference model, obsinfo and actinfo, respectively.

example

[___] = createintegratedenv(___,name=value) creates a model and environment interface using port, observation, and action sets specified using one or more name=value arguments.

examples

this example uses:

this example shows how to use createintegratedenv to create an environment object starting from a simulink model that implements the system with which the agent will interact, and that does not have an agent block. such a system is often referred to as plant, open-loop system, or reference system, while the whole (integrated) system that includes the agent is often referred to as the closed-loop system.

for this example, use the flying robot model described in train ddpg agent to control sliding robot as the reference (open-loop) system.

open the flying robot model.

open_system("rlflyingrobotenv")

initialize the state variables and sample time.

% initial model state variables
theta0 = 0;
x0 = -15;
y0 = 0;
% sample time
ts = 0.4;

create the simulink model myintegratedenv containing the flying robot model connected in a closed loop to the agent block. the function also returns the reinforcement learning environment object env to be used for training.

env = createintegratedenv( ...
    "rlflyingrobotenv", ...
    "myintegrateden")
env = 
simulinkenvwithagent with properties:
           model : myintegrateden
      agentblock : myintegrateden/rl agent
        resetfcn : []
  usefastrestart : on

the function can also return the block path to the rl agent block in the new integrated model, as well as the observation and action specifications for the reference model. 

[~,agentblk,observationinfo,actioninfo] = ...
    createintegratedenv( ...
    "rlflyingrobotenv","myintegratedenv")
agentblk = 
"myintegratedenv/rl agent"

observationinfo = 
  rlnumericspec with properties:
     lowerlimit: -inf
     upperlimit: inf
           name: "observation"
    description: [0x0 string]
      dimension: [7 1]
       datatype: "double"

actioninfo = 
  rlnumericspec with properties:
     lowerlimit: -inf
     upperlimit: inf
           name: "action"
    description: [0x0 string]
      dimension: [2 1]
       datatype: "double"

returning the block path and specifications is useful in cases in which you need to modify descriptions, limits, or names in observationinfo and actioninfo. after modifying the specifications, you can then create an environment from the integrated model integratedenv using the rlsimulinkenv function.

this example uses:

open the open-loop water tank model.

open_system("rlwatertankopenloop")

set the sample time of the discrete integrator block used to generate the observation, so the simulation can run.

ts = 1;

call createintegratedenv using name-value pairs to specify port names. the first argument of createintegratedenv is the name of the reference simulink model that contains the system with which the agent must interact. such a system is often referred to as plant, or open-loop system.

for this example, the reference system is the model of a water tank. the input port is called u (instead of action), and the first and third output ports are called y and stop (instead of observation and isdone). specify the port names using name-value pairs.

env = createintegratedenv("rlwatertankopenloop","integratedwatertank",...
    actionportname="u",observationportname="y",isdoneportname="stop")
env = 
simulinkenvwithagent with properties:
           model : integratedwatertank
      agentblock : integratedwatertank/rl agent
        resetfcn : []
  usefastrestart : on

the new model integratedwatertank contains the reference model connected in a closed-loop with the agent block. the function also returns the reinforcement learning environment object to be used for training.

input arguments

name of environment reference model, specified as a string or character vector. this is the simulink model implementing the environment the agent interacts with. such a system is often referred to as plant or open loop system, while the whole (integrated) system that includes both agent and environment is often referred to as the closed loop system. the generated model newmodel is a closed loop system that contains an rl agent block and references refmodel within its environment block. for more information on model referencing, see (simulink).

note

the reward signal at time t must be the one corresponding to the transition between the observation output at time t-1 and the observation output at time t. therefore, the environment output signal corresponds to the signal called next observation in the agent-environment illustration presented in reinforcement learning environments.

if your observation contains multiple channels, group the signals carried by the channels into a single observation bus. for more information about bus signals, see (simulink).

note

to avoid (potentially unsolvable) algebraic loops, you must avoid any direct feedthrough (that is any direct dependency in the same time step) from the action to any of the output signals. this is because simulink treats the agent block that is eventually added to the model as having a direct feedthrough from all its inputs (that is the action output at a given time step is considered to be directly dependent on the observation, reward and is-done inputs at the same time step). additionally, if the environment block is a referenced subsystem it is also normally treated as a direct feedthrough block unless the minimize algebraic loop occurrences parameter is enabled.

in general, adding a (simulink) or (simulink) block to the action signal between the agent and environment blocks removes the algebraic loop (alternatively you can add delay or memory blocks to all the environment output signals). for more information on algebraic loops and how to remove some of them, see (simulink) and (simulink).

name of the generated model, specified as a string or character vector. createintegratedenv creates a simulink model with this name, but does not save the model. the created model newmodel is a closed loop system that contains an rl agent block and references refmodel within its environment block. for more information on model referencing, see (simulink).

name-value arguments

specify optional pairs of arguments as name1=value1,...,namen=valuen, where name is the argument name and value is the corresponding value. name-value arguments must appear after other arguments, but the order of the pairs does not matter.

before r2021a, use commas to separate each name and value, and enclose name in quotes.

example: isdoneportname="stopsim" sets the stopsim port of the reference model as the source of the isdone signal.

name of the observation output port in the environment reference model, specified as a string or character vector. specify observationportname when the name of the observation output port of the reference model is not "observation".

example: observationportname="x"

name of the action input port in the environment reference model, specified as a string or character vector. specify actionportname when the name of the action input port of the reference model is not "action".

example: actionportname="u"

name of the reward output port in the environment reference model, specified as a string or character vector. specify rewardportname when the name of the reward output port of the reference model is not "reward".

example: rewardportname="r"

name of the is-done output port in the environment reference model, specified as a string or character vector. specify isdoneportname when the name of the is-done flag output port of the reference model is not "isdone".

example: isdoneportname="done"

names of observation bus leaf elements for which to create specifications, specified as a string array. to create observation specifications for a subset of the elements in a simulink bus object, specify buselementnames. if you do not specify buselementnames, a data specification is created for each leaf element in the bus.

observationbuselementnames is applicable only when the observation output port is a bus signal.

example: observationbuselementnames=["sin" "cos"] creates specifications for the observation bus elements with the names "sin" and "cos".

elements of finite observation set, specified as a cell array of name-value pairs. each name-value pair consists of an element name and an array of discrete values.

if the observation output port of the reference model is:

the specified discrete values must be castable to the data type of the observation signal arriving to the observation output port in the environment reference model.

if you do not specify discrete values for an observation channel, the signals carried by the channel are continuous.

example: observationdiscreteelements={"observation",[-1 0 1]} specifies discrete values for a nonbus observation signal with port name observation.

example: observationdiscreteelements={"gear",[-1 0 1 2],"direction",[1 2 3 4]} specifies discrete values for the "gear" and "direction" leaf elements of a bus action signal.

elements of finite action set, specified as a cell array of name-value pairs. each name-value pair consists of an element name and an array of discrete values.

if the action input port of the reference model is:

  • a bus signal, specify the name of a leaf element of the bus

  • nonbus signal, specify the name of the action port, as specified by actionportname

the specified discrete values must be castable to the data type of the action signal that can be accepted by the action input port in the environment reference model.

if you do not specify discrete values for the action channel, the signals carried by the channel are continuous.

example: actiondiscreteelements={"action",[-1 0 1]} specifies discrete values for a nonbus action signal with port name "action".

example: actiondiscretelements={"force",[-10 0 10],"torque",[-5 0 5]} specifies discrete values for the 'force' and 'torque' leaf elements of a bus action signal.

note

while creating an integrated environment more than action channel is possible, reinforcement learning toolbox™ agents only allow a single action channel.

output arguments

reinforcement learning environment interface, returned as an simulinkenvwithagent object. you can use this object to train and simulate agents in the same way as with any other environment.

for more information on reinforcement learning environments, see create custom simulink environments.

block path to the agent block in the new model, returned as a string. to train an agent in the new simulink model, you must create an agent and specify the agent name in the rl agent block indicated by agentblock.

for more information on creating agents, see reinforcement learning agents.

observation data specifications, returned as one of the following:

  • rlnumericspec object for a single continuous observation specification

  • rlfinitesetspec object for a single discrete observation specification

  • array of data specification objects for multiple specifications

action data specifications, returned as one of the following:

  • rlnumericspec object for a single continuous action specification

  • rlfinitesetspec object for a single discrete action specification

  • array of data specification objects for multiple action specifications

note

while creating an integrated environment more than action channel is possible, reinforcement learning toolbox agents only allow a single action channel.

version history

introduced in r2019a

see also

functions

objects

blocks

topics

网站地图