Error while training a two-agent environment: Action specification of agent at index 2 incompatible with environment

I am trying to simulate a two-agent system, using the following structure, and I am no using a reset function at all.

out = sim(envNoLA,[agentTray agentGains],simOpts);

But I get this error:

Error using rl.util.validateAgentsWithEnv>localValidateForSim

Action specification of agent at index 2 incompatible with environment.

Verify the action specification of the agents and environment using

getActionInfo.

Error in rl.util.validateAgentsWithEnv (line 23)

localValidateForSim(agents,env);

Error in rl.env.AbstractEnv/sim (line 67)

rl.util.validateAgentsWithEnv(policies,env);

After checking the specifications for each agent I received:

getActionInfo(agentTray)
    
    ans = 
    
      rlNumericSpec with properties:
    
         LowerLimit: -0.0349
         UpperLimit: 1.0472
               Name: "Trajectories"
        Description: [0×0 string]
          Dimension: [8 1]
           DataType: "double"

And

getActionInfo(agentGains)

ans = 

  rlNumericSpec with properties:

     LowerLimit: -2
     UpperLimit: 50
           Name: "Trajectories"
    Description: [0×0 string]
      Dimension: [24 1]
       DataType: "double"

The stablished options are:

simOpts = rlSimulationOptions('MaxSteps',1000,...
    'NumSimulations',3);

How can I correct this issue?

[郭欣]

have you solved the problem yet? I met the same question ....

Answer 1

[Emmanouil Tzorakoleftherakis]

The issue seems to be that the action specifications do not match what the environment model expects. Whether you are using Simulink or MATLAB to model the environment, make sure that the environment input matches what the agent outputs, primarily the dimensions of those actions