// Returns a list of float defining the state of the agent
public override List <float> CollectState()
{
List <float> state = new List <float>();
// @note: we handle changing state size (i.e. more marbles) by applying state size to the brain based on the number of marbles
// we get the bottomest marble only with negative velocity (i.e. the one that is going down!)
// this is because we cannot change state size with the curriculum, so we cannot just add more marbles and check all of them
// also, we are not interested in high marbles, after all
// @note: we could check the two bottomest marbles if needed
Marble riskyestMarble = marbles[0];
if (!stateUseAllMarbles)
{
for (var mi = 1; mi < nMarbles; mi++) // Only the first nMarbles are considered
{
var marble = marbles[mi];
var velOld = riskyestMarble.GetComponent <Rigidbody>().velocity.y;
var velNew = marble.GetComponent <Rigidbody>().velocity.y;
// TODO: cache RB into marble itself so we do not call GetComponent
if (velNew < velOld)
{
// going down-er!
riskyestMarble = marble;
}
else if (Math.Abs(velNew - velOld) < 0.01f)
{
if (marble.transform.position.y < riskyestMarble.transform.position.y)
{
riskyestMarble = marble;
}
}
}
foreach (var marble in marbles)
{
marble.GetComponentInChildren <MeshRenderer>().material.color = Color.white;
}
riskyestMarble.GetComponentInChildren <MeshRenderer>().material.color = Color.red;
}
for (var mi = 0; mi < nMarbles; mi++) // Only the first nMarbles are considered
{
var marble = marbles[mi];
var usedMarble = marble;
if (!stateUseAllMarbles)
{
usedMarble = riskyestMarble;
}
// Marble Position (+3)
// @note: normalized and relative to agent's position
if (stateUseMarblePosVelXZ)
{
state.Add((usedMarble.transform.position.x - transform.position.x) / (gridSize / 2f));
}
if (stateUseMarblePosVelY)
{
state.Add((usedMarble.transform.position.y - transform.position.y) / (areaHeight));
}
if (stateUseMarblePosVelXZ)
{
state.Add((usedMarble.transform.position.z - transform.position.z) / (gridSize / 2f));
}
// TODO: marbles should have a MAX SPEED
// Marble Speed (+3)
// @note: normalized based on max velocity
var marbleRB = usedMarble.transform.GetComponent <Rigidbody>();
var marbleVelocity = marbleRB.velocity;
if (stateUseMarblePosVelXZ)
{
state.Add(Mathf.Clamp(marbleVelocity.x, -maxMarbleVelocity, maxMarbleVelocity) / maxMarbleVelocity);
}
if (stateUseMarblePosVelY)
{
state.Add(Mathf.Clamp(marbleVelocity.y, -maxMarbleVelocity, maxMarbleVelocity) / maxMarbleVelocity);
}
if (stateUseMarblePosVelXZ)
{
state.Add(Mathf.Clamp(marbleVelocity.z, -maxMarbleVelocity, maxMarbleVelocity) / maxMarbleVelocity);
}
// TODO: marbles should have a MAX angular velocity
// Marble Angular Velocity (+3)
var marbleAngularVelocity = marbleRB.angularVelocity;
if (stateUseMarblePosVelXZ)
{
state.Add(Mathf.Clamp(marbleAngularVelocity.x, -maxMarbleVelocity, maxMarbleVelocity) /
maxMarbleVelocity);
}
if (stateUseMarblePosVelY)
{
state.Add(Mathf.Clamp(marbleAngularVelocity.y, -maxMarbleVelocity, maxMarbleVelocity) /
maxMarbleVelocity);
}
if (stateUseMarblePosVelXZ)
{
state.Add(Mathf.Clamp(marbleAngularVelocity.z, -maxMarbleVelocity, maxMarbleVelocity) /
maxMarbleVelocity);
}
if (!stateUseAllMarbles)
{
break; // Only one
}
}
foreach (var block in blocks)
{
// Block trigger time (+1)
if (stateUseBlockTriggerTime)
{
state.Add(block.CurrentTriggerTime);
}
// Block position (+2)
if (stateUseBlockPos)
{
state.Add((block.transform.position.x - transform.position.x) / (gridSize / 2f));
//state.Add((block.transform.position.y - transform.position.y) / (gridSize / 2f));
state.Add((block.transform.position.z - transform.position.z) / (gridSize / 2f));
}
// Block rotation (+3)
if (stateUseBlockRot)
{
state.Add(block.transform.localEulerAngles.x / blockRotationAngleLimit);
state.Add(block.transform.localEulerAngles.y / blockRotationAngleLimit);
state.Add(block.transform.localEulerAngles.z / blockRotationAngleLimit);
}
// Block velocity (+2)
if (stateUseBlockVel)
{
// TODO: range -1,1
//var blockRb = block.GetComponent<Rigidbody>();
//var blockVel = blockRb.velocity;
state.Add(Mathf.Clamp(block.speed.x, -maxBlockVelocity, maxBlockVelocity) / maxBlockVelocity);
//state.Add(Mathf.Clamp(blockVel.y, -maxBlockVelocity, maxBlockVelocity) / maxBlockVelocity);
state.Add(Mathf.Clamp(block.speed.z, -maxBlockVelocity, maxBlockVelocity) / maxBlockVelocity);
}
}
if (verboseState)
{
string statesString = "State:";
for (int i = 0; i < state.Count; i++)
{
statesString += " " + state[i] + "\n";
}
Debug.Log(statesString);
}
return(state);
}
