Simple number-picking ML agent code not working

Unity Engine
, , ,
1

I basically have an agent that does:

public override void OnActionReceived(ActionBuffers actionBuffers)
{
    if (Mathf.RoundToInt(actionBuffers.ContinuousActions[0]) < 0)
        SetReward(1f);
    else
        SetReward(-1f);
    EndEpisode();
}

I would assume this would learn to pick floats that round to -1. Instead it seems to do the opposite…

image
image1280×406 22.7 KB

Can someone explain what’s wrong?