I have a question about my learning.

I am doing the following learning.

If the square object is below, the agent is judged to be below, and if it is above, the agent is judged to be above.


I wrote the code like this:
public override void CollectObservations(VectorSensor sensor)
{
sensor.AddObservation(moveObject.getYpos());
sensor.AddObservation(actionY);
}

public override void OnActionReceived(float[] vectorAction)
{
actionY = Mathf.FloorToInt(vectorAction[0]);
float yPos = moveObject.getYpos();
gameObject.transform.position = new Vector3(gameObject.transform.position.x,
actionY + 1f, gameObject.transform.position.z);
if (actionY == 0)
{
if(yPos >= 1f)
{
AddReward(-1f);
}
else
{
AddReward(1f);
}
}
else if(actionY == 1)
{
if(yPos >= 1f)
{
AddReward(1f);
}
else
{
AddReward(-1f);
}
}
}


After that I tried to see the tensorboard, but the tensorboard did not turn on. Maybe tensorboard is not connected when connected by wifi? Should I connect only with LAN lines?

Since tensorboard does not work, I tried to create a graph directly from the csv file, but the result is normally returned except for Policy Loss.

Policy Loss is a form of continuous vibration.


Why is this appearance? Please answer.