Adds Platform2D

README.md

@@ -38,4 +38,9 @@ https://github.com/user-attachments/assets/80c8b16b-df09-4607-bcc6-2b0e760f03c5
 #### Robot FPS:
 https://github.com/user-attachments/assets/d44efbd1-59c2-4828-ae88-d8b374fb27e2
 
+#### Platform2D environment:
+https://github.com/user-attachments/assets/468f3eb5-ea9f-4eb0-8ca1-6b8b67f37d02
+
+
+
 

examples/Platform2D/.gitattributes

@@ -0,0 +1,2 @@
+# Normalize EOL for all files that Git considers text files.
+* text=auto eol=lf

examples/Platform2D/.gitignore

@@ -0,0 +1,3 @@
+# Godot 4+ specific ignores
+.godot/
+android/

examples/Platform2D/Platform2D.csproj

@@ -0,0 +1,11 @@
+<Project Sdk="Godot.NET.Sdk/4.3.0">
+  <PropertyGroup>
+    <TargetFramework>net6.0</TargetFramework>
+    <TargetFramework Condition=" '$(GodotTargetPlatform)' == 'android' ">net7.0</TargetFramework>
+    <TargetFramework Condition=" '$(GodotTargetPlatform)' == 'ios' ">net8.0</TargetFramework>
+    <EnableDynamicLoading>true</EnableDynamicLoading>
+  </PropertyGroup>
+  <ItemGroup>
+    <PackageReference Include="Microsoft.ML.OnnxRuntime" Version="1.15.1" />
+  </ItemGroup>
+</Project>

examples/Platform2D/Platform2D.sln

@@ -0,0 +1,19 @@
+Microsoft Visual Studio Solution File, Format Version 12.00
+# Visual Studio 2012
+Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "Platform2D", "Platform2D.csproj", "{8552EC7B-EF81-42D4-828B-B6CD9D17C897}"
+EndProject
+Global
+	GlobalSection(SolutionConfigurationPlatforms) = preSolution
+	Debug|Any CPU = Debug|Any CPU
+	ExportDebug|Any CPU = ExportDebug|Any CPU
+	ExportRelease|Any CPU = ExportRelease|Any CPU
+	EndGlobalSection
+	GlobalSection(ProjectConfigurationPlatforms) = postSolution
+		{8552EC7B-EF81-42D4-828B-B6CD9D17C897}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
+		{8552EC7B-EF81-42D4-828B-B6CD9D17C897}.Debug|Any CPU.Build.0 = Debug|Any CPU
+		{8552EC7B-EF81-42D4-828B-B6CD9D17C897}.ExportDebug|Any CPU.ActiveCfg = ExportDebug|Any CPU
+		{8552EC7B-EF81-42D4-828B-B6CD9D17C897}.ExportDebug|Any CPU.Build.0 = ExportDebug|Any CPU
+		{8552EC7B-EF81-42D4-828B-B6CD9D17C897}.ExportRelease|Any CPU.ActiveCfg = ExportRelease|Any CPU
+		{8552EC7B-EF81-42D4-828B-B6CD9D17C897}.ExportRelease|Any CPU.Build.0 = ExportRelease|Any CPU
+	EndGlobalSection
+EndGlobal

examples/Platform2D/addons/godot_rl_agents/controller/ai_controller_2d.gd

@@ -0,0 +1,136 @@
+extends Node2D
+class_name AIController2D
+
+enum ControlModes {
+	INHERIT_FROM_SYNC, ## Inherit setting from sync node
+	HUMAN, ## Test the environment manually
+	TRAINING, ## Train a model
+	ONNX_INFERENCE, ## Load a pretrained model using an .onnx file
+	RECORD_EXPERT_DEMOS ## Record observations and actions for expert demonstrations
+}
+@export var control_mode: ControlModes = ControlModes.INHERIT_FROM_SYNC
+## The path to a trained .onnx model file to use for inference (overrides the path set in sync node).
+@export var onnx_model_path := ""
+## Once the number of steps has passed, the flag 'needs_reset' will be set to 'true' for this instance.
+@export var reset_after := 1000
+
+@export_group("Record expert demos mode options")
+## Path where the demos will be saved. The file can later be used for imitation learning.
+@export var expert_demo_save_path: String
+## The action that erases the last recorded episode from the currently recorded data.
+@export var remove_last_episode_key: InputEvent
+## Action will be repeated for n frames. Will introduce control lag if larger than 1.
+## Can be used to ensure that action_repeat on inference and training matches
+## the recorded demonstrations.
+@export var action_repeat: int = 1
+
+@export_group("Multi-policy mode options")
+## Allows you to set certain agents to use different policies.
+## Changing has no effect with default SB3 training. Works with Rllib example.
+## Tutorial: https://github.com/edbeeching/godot_rl_agents/blob/main/docs/TRAINING_MULTIPLE_POLICIES.md
+@export var policy_name: String = "shared_policy"
+
+var onnx_model: ONNXModel
+
+var heuristic := "human"
+var done := false
+var reward := 0.0
+var n_steps := 0
+var needs_reset := false
+
+var _player: Node2D
+
+
+func _ready():
+	add_to_group("AGENT")
+
+
+func init(player: Node2D):
+	_player = player
+
+
+#region Methods that need implementing using the "extend script" option in Godot
+func get_obs() -> Dictionary:
+	assert(false, "the get_obs method is not implemented when extending from ai_controller")
+	return {"obs": []}
+
+
+func get_reward() -> float:
+	assert(false, "the get_reward method is not implemented when extending from ai_controller")
+	return 0.0
+
+
+func get_action_space() -> Dictionary:
+	assert(
+		false, "the get_action_space method is not implemented when extending from ai_controller"
+	)
+	return {
+		"example_actions_continous": {"size": 2, "action_type": "continuous"},
+		"example_actions_discrete": {"size": 2, "action_type": "discrete"},
+	}
+
+
+func set_action(action) -> void:
+	assert(false, "the set_action method is not implemented when extending from ai_controller")
+
+
+#endregion
+
+
+#region Methods that sometimes need implementing using the "extend script" option in Godot
+# Only needed if you are recording expert demos with this AIController
+func get_action() -> Array:
+	assert(
+		false,
+		"the get_action method is not implemented in extended AIController but demo_recorder is used"
+	)
+	return []
+
+
+# For providing additional info (e.g. `is_success` for SB3 training)
+func get_info() -> Dictionary:
+	return {}
+
+
+#endregion
+
+
+func _physics_process(delta):
+	n_steps += 1
+	if n_steps > reset_after:
+		needs_reset = true
+
+
+func get_obs_space():
+	# may need overriding if the obs space is complex
+	var obs = get_obs()
+	return {
+		"obs": {"size": [len(obs["obs"])], "space": "box"},
+	}
+
+
+func reset():
+	n_steps = 0
+	needs_reset = false
+
+
+func reset_if_done():
+	if done:
+		reset()
+
+
+func set_heuristic(h):
+	# sets the heuristic from "human" or "model" nothing to change here
+	heuristic = h
+
+
+func get_done():
+	return done
+
+
+func set_done_false():
+	done = false
+
+
+func zero_reward():
+	reward = 0.0

examples/Platform2D/addons/godot_rl_agents/controller/ai_controller_3d.gd

@@ -0,0 +1,136 @@
+extends Node3D
+class_name AIController3D
+
+enum ControlModes {
+	INHERIT_FROM_SYNC, ## Inherit setting from sync node
+	HUMAN, ## Test the environment manually
+	TRAINING, ## Train a model
+	ONNX_INFERENCE, ## Load a pretrained model using an .onnx file
+	RECORD_EXPERT_DEMOS ## Record observations and actions for expert demonstrations
+}
+@export var control_mode: ControlModes = ControlModes.INHERIT_FROM_SYNC
+## The path to a trained .onnx model file to use for inference (overrides the path set in sync node).
+@export var onnx_model_path := ""
+## Once the number of steps has passed, the flag 'needs_reset' will be set to 'true' for this instance.
+@export var reset_after := 1000
+
+@export_group("Record expert demos mode options")
+## Path where the demos will be saved. The file can later be used for imitation learning.
+@export var expert_demo_save_path: String
+## The action that erases the last recorded episode from the currently recorded data.
+@export var remove_last_episode_key: InputEvent
+## Action will be repeated for n frames. Will introduce control lag if larger than 1.
+## Can be used to ensure that action_repeat on inference and training matches
+## the recorded demonstrations.
+@export var action_repeat: int = 1
+
+@export_group("Multi-policy mode options")
+## Allows you to set certain agents to use different policies.
+## Changing has no effect with default SB3 training. Works with Rllib example.
+## Tutorial: https://github.com/edbeeching/godot_rl_agents/blob/main/docs/TRAINING_MULTIPLE_POLICIES.md
+@export var policy_name: String = "shared_policy"
+
+var onnx_model: ONNXModel
+
+var heuristic := "human"
+var done := false
+var reward := 0.0
+var n_steps := 0
+var needs_reset := false
+
+var _player: Node3D
+
+
+func _ready():
+	add_to_group("AGENT")
+
+
+func init(player: Node3D):
+	_player = player
+
+
+#region Methods that need implementing using the "extend script" option in Godot
+func get_obs() -> Dictionary:
+	assert(false, "the get_obs method is not implemented when extending from ai_controller")
+	return {"obs": []}
+
+
+func get_reward() -> float:
+	assert(false, "the get_reward method is not implemented when extending from ai_controller")
+	return 0.0
+
+
+func get_action_space() -> Dictionary:
+	assert(
+		false, "the get_action_space method is not implemented when extending from ai_controller"
+	)
+	return {
+		"example_actions_continous": {"size": 2, "action_type": "continuous"},
+		"example_actions_discrete": {"size": 2, "action_type": "discrete"},
+	}
+
+
+func set_action(action) -> void:
+	assert(false, "the set_action method is not implemented when extending from ai_controller")
+
+
+#endregion
+
+
+#region Methods that sometimes need implementing using the "extend script" option in Godot
+# Only needed if you are recording expert demos with this AIController
+func get_action() -> Array:
+	assert(
+		false,
+		"the get_action method is not implemented in extended AIController but demo_recorder is used"
+	)
+	return []
+
+
+# For providing additional info (e.g. `is_success` for SB3 training)
+func get_info() -> Dictionary:
+	return {}
+
+
+#endregion
+
+
+func _physics_process(delta):
+	n_steps += 1
+	if n_steps > reset_after:
+		needs_reset = true
+
+
+func get_obs_space():
+	# may need overriding if the obs space is complex
+	var obs = get_obs()
+	return {
+		"obs": {"size": [len(obs["obs"])], "space": "box"},
+	}
+
+
+func reset():
+	n_steps = 0
+	needs_reset = false
+
+
+func reset_if_done():
+	if done:
+		reset()
+
+
+func set_heuristic(h):
+	# sets the heuristic from "human" or "model" nothing to change here
+	heuristic = h
+
+
+func get_done():
+	return done
+
+
+func set_done_false():
+	done = false
+
+
+func zero_reward():
+	reward = 0.0

examples/Platform2D/addons/godot_rl_agents/godot_rl_agents.gd

@@ -0,0 +1,16 @@
+@tool
+extends EditorPlugin
+
+
+func _enter_tree():
+	# Initialization of the plugin goes here.
+	# Add the new type with a name, a parent type, a script and an icon.
+	add_custom_type("Sync", "Node", preload("sync.gd"), preload("icon.png"))
+	#add_custom_type("RaycastSensor2D2", "Node", preload("raycast_sensor_2d.gd"), preload("icon.png"))
+
+
+func _exit_tree():
+	# Clean-up of the plugin goes here.
+	# Always remember to remove it from the engine when deactivated.
+	remove_custom_type("Sync")
+	#remove_custom_type("RaycastSensor2D2")