|
extends AIController3D |
|
|
|
|
|
@onready var grid_sensor_3d = $GridSensor3D |
|
|
|
var fb_action : float = 0.0 |
|
var turn_action : float = 0.0 |
|
var _player_positions : = Deque.new() |
|
|
|
|
|
func _ready(): |
|
super._ready() |
|
_player_positions.max_size = 80 # 10 steps with repeat of 8 |
|
|
|
func get_obs() -> Dictionary: |
|
var obs = grid_sensor_3d.get_observation() |
|
return {"obs": obs} |
|
|
|
func get_reward() -> float: |
|
return reward + keep_moving_reward() |
|
|
|
func keep_moving_reward() -> float: |
|
var mean_position = Vector3.ZERO |
|
for pos in _player_positions: |
|
mean_position += pos |
|
mean_position /= _player_positions.size() |
|
|
|
var penalty = 0.0 |
|
|
|
for pos in _player_positions: |
|
penalty += (pos-mean_position).length() |
|
|
|
#prints("penalty", 0.01*(penalty / _player_positions.size())) |
|
|
|
return 0.005*(penalty / _player_positions.size()) |
|
|
|
func _physics_process(delta): |
|
_player_positions.push_back(_player.position) |
|
|
|
func reset(): |
|
super.reset() |
|
_player_positions.clear() |
|
_player_positions.push_back(_player.position) |
|
|
|
|
|
|
|
func get_action_space() -> Dictionary: |
|
return { |
|
"fb" : { |
|
"size": 1, |
|
"action_type": "continuous" |
|
}, |
|
"turn" : { |
|
"size": 1, |
|
"action_type": "continuous" |
|
}, |
|
} |
|
|
|
func set_action(action) -> void: |
|
fb_action = clamp(action["fb"][0], -1.0, 1.0) |
|
turn_action = clamp(action["turn"][0], -1.0, 1.0) |
|
|