Bug Instantiate Agent with GridSensor

Hi. I'm trying to runtime add (instantiate) agent (prefabe) to scene. But I get the error:

Spoiler: error

NullReferenceException: Object reference not set to an instance of an object
Unity.MLAgents.Extensions.Sensors.GridSensor.ChannelsToTexture (System.Int32 channelIndex, System.Int32 numChannelsToAdd) (at C:/Users/User/ml-agents/com.unity.ml-agents.extensions/Runtime/Sensors/GridSensor.cs:548)
Unity.MLAgents.Extensions.Sensors.GridSensor.GetCompressedObservation () (at C:/Users/User/ml-agents/com.unity.ml-agents.extensions/Runtime/Sensors/GridSensor.cs:523)
Unity.MLAgents.GrpcExtensions.GetObservationProto (Unity.MLAgents.Sensors.ISensor sensor, Unity.MLAgents.Sensors.ObservationWriter observationWriter) (at C:/Users/User/ml-agents/com.unity.ml-agents/Runtime/Communicator/GrpcExtensions.cs:406)
Unity.MLAgents.RpcCommunicator.PutObservations (System.String behaviorName, Unity.MLAgents.AgentInfo info, System.Collections.Generic.List`1[T] sensors) (at C:/Users/User/ml-agents/com.unity.ml-agents/Runtime/Communicator/RpcCommunicator.cs:330)
Unity.MLAgents.Policies.RemotePolicy.RequestDecision (Unity.MLAgents.AgentInfo info, System.Collections.Generic.List`1[T] sensors) (at C:/Users/User/ml-agents/com.unity.ml-agents/Runtime/Policies/RemotePolicy.cs:55)
Unity.MLAgents.Agent.SendInfoToBrain () (at C:/Users/User/ml-agents/com.unity.ml-agents/Runtime/Agent.cs:1126)
Unity.MLAgents.Agent.SendInfo () (at C:/Users/User/ml-agents/com.unity.ml-agents/Runtime/Agent.cs:1387)
Unity.MLAgents.Academy.EnvironmentStep () (at C:/Users/User/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:573)
Unity.MLAgents.AcademyFixedUpdateStepper.FixedUpdate () (at C:/Users/User/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:43)

it happens everytime only once for the entire learning game, regardless of the number of agents I'm trying to create. But everything works. But it's still strange. and I don't understand why this is happening.

Spoiler: Version information:

CMD: Version information:
ml-agents: 0.25.0,
ml-agents-envs: 0.25.0,
Communicator API: 1.5.0,
PyTorch: 1.7.1+cu110

Ver:
ML Agents 1.9.0-preview
Unity 2021.1.1f1

 

In new version.
When I set the "Observation Stacks" more than 1, the error appears:

Spoiler: error

InvalidCastException: Specified cast is not valid.
Unity.MLAgents.Sensors.GridSensorComponent.CreateSensors () (at D:/Unity/ml-agents-release_17_branch/com.unity.ml-agents/Runtime/Sensors/GridSensorComponent.cs:237)
Unity.MLAgents.Agent.InitializeSensors () (at D:/Unity/ml-agents-release_17_branch/com.unity.ml-agents/Runtime/Agent.cs:977)
Unity.MLAgents.Editor.BehaviorParametersEditor.DisplayFailedModelChecks () (at D:/Unity/ml-agents-release_17_branch/com.unity.ml-agents/Editor/BehaviorParametersEditor.cs:110)
Unity.MLAgents.Editor.BehaviorParametersEditor.OnInspectorGUI () (at D:/Unity/ml-agents-release_17_branch/com.unity.ml-agents/Editor/BehaviorParametersEditor.cs:81)
UnityEditor.UIElements.InspectorElement+<>c__DisplayClass59_0.<CreateIMGUIInspectorFromEditor>b__0 () (at <a122028f8c314654ad750c3d9eb1f725>:0)
UnityEngine.GUIUtilityrocessEvent(Int32, IntPtr, Boolean&)

 

New version it work fine in my project (by create runtime agents)!

 

But strange. earlier warning "Couldn't connect to trainer on port 5004 using ..." appeared before start scene, but in new ver, after.

 

is simple.
take a clean Release 15 ML-Agents.
Go Example -> FoodCollector -> FoodCollectorSettings -> EnvironmentReset
and write bellow in the end:

Code (CSharp):

1. foreach (var a in agents)
2.         {
3.             Instantiate(a,a.transform.parent);
4.         }

and look error NullReferenceException.
it comes out only once at the beginning and does not affect anything. everything works correctly.

 

Sorry. This message appears when there is at least one agent in the scene. And since the new version can easily create an agent from the prefab, I dont need to leave them in scene, and the initialization occurs later. Everything is correct. Thank you.