Movement jitter

[jeremyhcho]

Apologies for bombarding the Q&A today but I've seemingly got things to work except interpolation and I'm not sure what I'm missing.

Here's my RollbackSynchronizer

My TickInterpolator

And my _rollback_tick code

func _rollback_tick(delta: float, _tick: int, _is_fresh: bool) -> void:
    # Add the gravity.
    _force_update_is_on_floor()
    if not is_on_floor():
        velocity.y -= gravity * delta

    var direction = Vector3(input.input_dir.x, 0, input.input_dir.y).normalized()
    if direction:
        velocity.x = direction.x * speed
        velocity.z = direction.z * speed
    else:
        velocity.x = move_toward(velocity.x, 0, speed)
        velocity.z = move_toward(velocity.z, 0, speed)

    # Apply movement
    velocity *= NetworkTime.physics_factor
    move_and_slide()
    velocity /= NetworkTime.physics_factor

With this setup I can move my character on the appropriate clients as well as see the correct characters move on opposite clients but the model jitters quite heavily during movement. According to the docs (unless I missed something right in front of me), this setup seems reasonable and would indicate that I should see smooth movement. Eyeball test indicates that the TickInterpolator does something because when I turn off the Enabled flag it feels different but it's somewhat hard to tell.

I would appreciate a nudge in the right direction! Thanks in advance!

[elementbound]

Hi @jeremyhcho!

This kind of jitter usually happens when there's a difference in what the client predicts and what the server responds with. The jitter itself is the client adjusting its state, which is usually not much of an issue, unless a lot of adjustment is needed.

Would you be able to check if input is properly received on the server? For example that the Input node's values approximately correspond to what you're pressing on the keyboard.

Also, would you be able to share a minimum reproducing project? I'd be happy to look into it. Otherwise your setup looks good.

Eyeball test indicates that the TickInterpolator does something

Protip: you can test and see the TickInterpolator in action by setting the tickrate to something really low, like 8 or 4.

[jeremyhcho]

Following up here. Any other suspicions as to what might be causing this? cc @elementbound

I've been experimenting with the Sync to physics option checked but this sets the NetworkTime.tickrate to 60 which is higher than I'd need. On top of that, when I have this checked, there's something that seems to be a race condition with my player input component that looks like

[WRN][netfox::RollbackSynchronizer] Input:is_jumping
[WRN][netfox::RollbackSynchronizer] Received input for node owned by 2135926936 from 1, sender has no authority!

I'd like to try and fix the jitter without checking Sync to physics. Here's the current state of things:

Player scene tree

RollbackSynchronizer

TickInterpolator

Player.ready()

func _ready() -> void:
	# Prevent the spring from detecting the character model as an obstacle
	spring.add_excluded_object(self)

	global_position = GameStart.DEFAULT_SPAWN
	yaw.global_position = Vector3(
		global_position.x,
		yaw.global_position.y,
		global_position.z
	)

	# Process a frame before processing settings
	await get_tree().process_frame

	# Set authorities
	set_multiplayer_authority(Constants.SERVER_MULTIPLAYER_ID)
	player_input.set_multiplayer_authority(peer_id)
	rollback_synchronizer.process_authority()

Player._rollback_tick()

func _rollback_tick(delta: float, _tick: int, _is_fresh: bool) -> void:
	_force_update_is_on_floor()

	yaw.rotation_degrees.y = player_input.yaw_value
	pitch.rotation_degrees.x = player_input.pitch_value

	# Rotate character to yaw
	if player_input.is_rotating and player_input.input_dir != Vector2.ZERO:
		rotation.y = yaw.rotation.y

	var direction = (basis * Vector3(player_input.input_dir.x, 0, player_input.input_dir.y)).normalized()
	speed = RUNNING_SPEED if player_input.input_dir.y <= 0 else WALKING_SPEED
	if is_on_floor():
		set_grounded_velocity(delta, direction)
		if player_input.is_jumping:
			jump()
	else:
		set_airborne_velocity(delta, direction)

	# Apply movement
	velocity *= NetworkTime.physics_factor
	move_and_slide()
	velocity /= NetworkTime.physics_factor

	yaw.global_position = Vector3(
		global_position.x,
		global_position.y + 1.5,
		global_position.z
	)

func set_grounded_velocity(_delta: float, direction: Vector3) -> void:
	if direction:
		velocity.x = direction.x * speed
		velocity.z = direction.z * speed
	else:
		velocity.x = move_toward(velocity.x, 0, speed)
		velocity.z = move_toward(velocity.z, 0, speed)

Player._force_update_is_on_floor()

func _force_update_is_on_floor() -> void:
	var old_velocity = velocity
	velocity *= 0
	move_and_slide()
	velocity = old_velocity

I've confirmed that I have no _physics_process function defined and that my _process function simply deals with animation blending but here's what _process looks like for posterity

func _process(delta: float) -> void:
	var animation_tree = assassin_model.animation_tree
	var input_dir = player_input.input_dir
	var iwr_blend = animation_tree.get("parameters/iwr_blend/blend_amount")

	# Blend locomotion
	animation_vector = animation_vector.lerp(input_dir, 0.2)
	if input_dir == Vector2.ZERO:
		animation_tree.set("parameters/iwr_blend/blend_amount", lerpf(iwr_blend, -1.0, 1 - exp(-delta * 10.0)))
	elif input_dir.y <= 0:
		animation_tree.set("parameters/run/blend_position", animation_vector)
		animation_tree.set("parameters/iwr_blend/blend_amount", lerpf(iwr_blend, 1.0, 1 - exp(-delta * 10.0)))
	else:
		animation_tree.set("parameters/walk/blend_position", animation_vector)
		animation_tree.set("parameters/iwr_blend/blend_amount", lerpf(iwr_blend, 0.0, 1 - exp(-delta * 10.0)))

	# Handle blend for is_on_floor
	var on_floor_blend_target = 1.0 if is_on_floor() else 0.0
	var on_floor_blend = animation_tree.get("parameters/on_floor_blend/blend_amount")
	animation_tree.set(
		"parameters/on_floor_blend/blend_amount",
		lerpf(on_floor_blend, on_floor_blend_target, 10 * delta)
	)

	if player_input.is_stabbing:
		if player_input.input_dir == Vector2.ZERO:
			assassin_model.animation_player.play("stab")
		else:
			assassin_model.animation_tree.set(
				"parameters/stab/request",
				AnimationNodeOneShot.ONE_SHOT_REQUEST_FIRE
			)

Not sure exactly how else to look into this. Any guidance would be very appreciated. Thanks!

[elementbound]

Hi @jeremyhcho! I've been busier than usual recently, but I'll try to take a deeper look in the following days.

In the meanwhile, if it doesn't conflict with other systems in your game, you can reduce the physics tickrate, so sync to physics can be active without causing too much network traffic.

In the meanwhile, if you could attach a Godot project that replicates this issue, I'd be happy to look at that, it would probably simplify debugging the issue.

[jeremyhcho]

No worries! Thanks for the offer.

The interesting bit is that I can't seem to reproduce this issue on a new project but when I slim my current project down to what I think is pretty bare bones I still see this issue. I have a private repo that I don't mind inviting you to for debugging if you're open to helping me out

[elementbound]

For sure @jeremyhcho, I'd be happy to look at the repo!

[elementbound]

Hey @jeremyhcho! I'm still down to debug your repo if you'd like. Don't want to nag, just checking in 🙂