config/marathon_envs_config.yaml + two minor fixes (#9)

* add training config file * info for marathon_envs_config.yaml * Fix: academy script references are broken. Fixed by changing guid to MujucoAacademy.cs * Fix: dm_Walker was terminating as 'right_foot', 'left_foot' where missing
Unity-Technologies · Sep 12, 2018 · 1de7098 · 1de7098
1 parent 6a30414
commit 1de7098
Show file tree

Hide file tree

Showing 8 changed files with 109 additions and 5 deletions.
diff --git a/MarathonEnvs/Environments/DeepMindHopper/Scenes/DeepMindHopper.unity b/MarathonEnvs/Environments/DeepMindHopper/Scenes/DeepMindHopper.unity
@@ -926,7 +926,7 @@ MonoBehaviour:
   m_GameObject: {fileID: 1574236047}
   m_Enabled: 1
   m_EditorHideFlags: 0
-  m_Script: {fileID: 11500000, guid: 9af83cd96d4bc4088a966af174446d1b, type: 3}
+  m_Script: {fileID: 11500000, guid: 4479159a705f245c3946cee6d1eb53ab, type: 3}
   m_Name: 
   m_EditorClassIdentifier: 
   maxSteps: 0

diff --git a/MarathonEnvs/Environments/DeepMindHumanoid/Scenes/DeepMindHumanoid.unity b/MarathonEnvs/Environments/DeepMindHumanoid/Scenes/DeepMindHumanoid.unity
@@ -1050,7 +1050,7 @@ MonoBehaviour:
   m_GameObject: {fileID: 1574236047}
   m_Enabled: 1
   m_EditorHideFlags: 0
-  m_Script: {fileID: 11500000, guid: 9af83cd96d4bc4088a966af174446d1b, type: 3}
+  m_Script: {fileID: 11500000, guid: 4479159a705f245c3946cee6d1eb53ab, type: 3}
   m_Name: 
   m_EditorClassIdentifier: 
   maxSteps: 0

diff --git a/MarathonEnvs/Environments/DeepMindWalker/Scenes/DeepMindWalker.unity b/MarathonEnvs/Environments/DeepMindWalker/Scenes/DeepMindWalker.unity
@@ -1022,7 +1022,7 @@ MonoBehaviour:
   m_GameObject: {fileID: 1574236047}
   m_Enabled: 1
   m_EditorHideFlags: 0
-  m_Script: {fileID: 11500000, guid: 9af83cd96d4bc4088a966af174446d1b, type: 3}
+  m_Script: {fileID: 11500000, guid: 4479159a705f245c3946cee6d1eb53ab, type: 3}
   m_Name: 
   m_EditorClassIdentifier: 
   maxSteps: 0

diff --git a/MarathonEnvs/Environments/OpenAIAnt/Scenes/OpenAIAnt.unity b/MarathonEnvs/Environments/OpenAIAnt/Scenes/OpenAIAnt.unity
@@ -1058,7 +1058,7 @@ MonoBehaviour:
   m_GameObject: {fileID: 1574236047}
   m_Enabled: 1
   m_EditorHideFlags: 0
-  m_Script: {fileID: 11500000, guid: 9af83cd96d4bc4088a966af174446d1b, type: 3}
+  m_Script: {fileID: 11500000, guid: 4479159a705f245c3946cee6d1eb53ab, type: 3}
   m_Name: 
   m_EditorClassIdentifier: 
   maxSteps: 0

diff --git a/MarathonEnvs/Environments/TestBed/Scenes/MarathonTestBed.unity b/MarathonEnvs/Environments/TestBed/Scenes/MarathonTestBed.unity
@@ -388,7 +388,7 @@ MonoBehaviour:
   m_GameObject: {fileID: 1574236047}
   m_Enabled: 1
   m_EditorHideFlags: 0
-  m_Script: {fileID: 11500000, guid: 9af83cd96d4bc4088a966af174446d1b, type: 3}
+  m_Script: {fileID: 11500000, guid: 4479159a705f245c3946cee6d1eb53ab, type: 3}
   m_Name: 
   m_EditorClassIdentifier: 
   maxSteps: 0

diff --git a/MarathonEnvs/Scripts/MarathonAgent.cs b/MarathonEnvs/Scripts/MarathonAgent.cs
@@ -439,6 +439,8 @@ public virtual void OnTerrainCollision(GameObject other, GameObject terrain)
                 case "right_ankle_geom": // oai_ant
                 case "third_ankle_geom": // oai_ant
                 case "fourth_ankle_geom": // oai_ant
+                case "right_foot": // dm_walker
+                case "left_foot": // dm_walker
                     FootHitTerrain = true;
                     break;
                 default:

diff --git a/README.md b/README.md
@@ -22,6 +22,7 @@ A set of high-dimensional continuous control environments for use with Unity ML-
  * Clone [ml-agents repository](https://github.com/Unity-Technologies/ml-agents).
  * Install [ML-Agents Toolkit](https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Installation.md).
  * Add `MarathonEnvs` sub-folder from this repository to `MLAgentsSDK\Assets\` in cloned ml-agents repository.
+ * Add `config\marathon_envs_config.yaml` from this reprository to `config\` in cloned ml-agents repository.
 
 ---
 ## Publications & Usage 
@@ -173,6 +174,8 @@ An early version of this work was presented March 19th, 2018 at the AI Summit -
     * DeepMindHopper - Folder for reproducing DeepMindHopper 
     * OpenAIAnt - Folder for reproducing OpenAIAnt 
     * etc
+* config
+  * marathon_envs_config.yaml - trainer-config file. The hyperparameters used when training from python.
 
 ### Tuning params / Magic numbers
 * xxNamexx\Prefab\xxNamexx -> MarathonSpawner.Force2D = set to True when implementing a 2d model (hopper, walker)

diff --git a/config/marathon_envs_config.yaml b/config/marathon_envs_config.yaml
@@ -0,0 +1,99 @@
+default:
+    trainer: ppo
+    batch_size: 1024
+    beta: 5.0e-3
+    buffer_size: 10240
+    epsilon: 0.2
+    gamma: 0.99
+    hidden_units: 128
+    lambd: 0.95
+    learning_rate: 3.0e-4
+    max_steps: 5.0e4
+    memory_size: 256
+    normalize: false
+    num_epoch: 3
+    num_layers: 2
+    time_horizon: 64
+    sequence_length: 64
+    summary_freq: 1000
+    use_recurrent: false
+    use_curiosity: false
+    curiosity_strength: 0.01
+    curiosity_enc_size: 128
+
+DeepMindHumanoidBrain:
+    normalize: true
+    num_epoch: 3
+    beta: 0.01
+    time_horizon: 1000
+    batch_size: 2048
+    buffer_size: 20480
+    gamma: 0.995
+    max_steps: 2e6
+    summary_freq: 1000
+    num_layers: 2
+    hidden_units: 512
+    # use_curiosity: true
+    # curiosity_strength: 0.01
+    # curiosity_enc_size: 256    
+
+DeepMindHopperBrain:
+    beta: 1.0e-2
+    epsilon: 0.20
+    gamma: 0.99
+    lambd: 0.95
+    learning_rate: 1.0e-3
+    num_epoch: 3
+    time_horizon: 128
+    summary_freq: 1000
+    use_recurrent: false
+    normalize: true
+    num_layers: 2
+    hidden_units: 90
+    batch_size: 2048
+    buffer_size: 10240
+    max_steps: 3e5
+    use_curiosity: true
+    curiosity_strength: 0.01
+    curiosity_enc_size: 256    
+
+DeepMindWalkerBrain:
+    beta: 1.0e-2
+    epsilon: 0.20
+    gamma: 0.99
+    lambd: 0.95
+    learning_rate: 1.0e-3
+    num_epoch: 3
+    time_horizon: 128
+    summary_freq: 1000
+    use_recurrent: false
+    normalize: true
+    num_layers: 3
+    hidden_units: 41
+    batch_size: 2048
+    buffer_size: 10240
+    max_steps: 3e5
+    use_curiosity: true
+    curiosity_strength: 0.01
+    curiosity_enc_size: 256    
+
+OpenAIAntBrain:
+    # beta: 1.0e-2
+    beta: 5.0e-3
+    epsilon: 0.20
+    gamma: 0.99
+    lambd: 0.95
+    learning_rate: 1.0e-3
+    num_epoch: 3
+    time_horizon: 128
+    summary_freq: 1000
+    use_recurrent: false
+    normalize: true
+    batch_size: 2048
+    buffer_size: 10240
+    num_layers: 3
+    hidden_units: 53
+    max_steps: 3e5
+    # use_curiosity: true
+    # curiosity_strength: 0.01
+    # curiosity_enc_size: 256