Merge branch 'master' into hh/develop/crawler-ragdoll-updates

DevProject/Packages/manifest.json

   "dependencies": {
     "com.unity.2d.sprite": "1.0.0",
     "com.unity.2d.tilemap": "1.0.0",
-    "com.unity.ads": "2.0.8",
+    "com.unity.ads": "3.4.4",
     "com.unity.analytics": "3.3.5",
     "com.unity.coding": "0.1.0-preview.13",
     "com.unity.collab-proxy": "1.2.16",
     "com.unity.multiplayer-hlapi": "1.0.4",
     "com.unity.package-manager-doctools": "1.1.1-preview.3",
-    "com.unity.package-validation-suite": "0.7.15-preview",
+    "com.unity.package-validation-suite": "0.11.0-preview",
-    "com.unity.test-framework": "1.1.11",
+    "com.unity.test-framework": "1.1.13",
-    "com.unity.xr.legacyinputhelpers": "1.3.8",
+    "com.unity.xr.legacyinputhelpers": "1.3.11",
     "com.unity.modules.ai": "1.0.0",
     "com.unity.modules.androidjni": "1.0.0",
     "com.unity.modules.animation": "1.0.0",

DevProject/ProjectSettings/ProjectVersion.txt

-m_EditorVersion: 2019.3.3f1
-m_EditorVersionWithRevision: 2019.3.3f1 (7ceaae5f7503)
+m_EditorVersion: 2019.3.11f1
+m_EditorVersionWithRevision: 2019.3.11f1 (ceef2d848e70)

Project/Assets/ML-Agents/Examples/Basic/TFModels/Basic.nn

-vector_observation����

action_masks����

policy_1/concat_2/concatactionaction_output_shape������?action_output_shape




memory_size





version_number




is_continuous_control




policy_1/add_2/y




policy_1/Sum/reduction_indices




policy_1/add/y




$policy/main_graph_0/hidden_0/BiasAdd
�����?
vector_observation#policy/main_graph_0/hidden_0/kernel

�
!policy/main_graph_0/hidden_0/bias


�
 policy/main_graph_0/hidden_0/Mul2	�����?
$policy/main_graph_0/hidden_0/BiasAddpolicy_1/dense/MatMul
�����?
 policy/main_graph_0/hidden_0/Mulpolicy/dense/kernel

�
<policy_1/dense/MatMul/patch:0


�
policy_1/strided_slice�



�����?
action_maskspolicy_1/Softmax2�����?
policy_1/dense/MatMulpolicy_1/addd�����?policy_1/Softmaxpolicy_1/add/ypolicy_1/Mulf�����?policy_1/addpolicy_1/strided_slicepolicy_1/Sum������?policy_1/Mulpolicy_1/Sum/reduction_indicespolicy_1/truedivg�����?policy_1/Mulpolicy_1/Sumpolicy_1/add_2d�����?policy_1/truedivpolicy_1/add_2/ypolicy_1/Log_12r�����?
policy_1/add_2policy_1/concat_2/concat2�����?
policy_1/Log_1action2�����?
policy_1/concat_2/concat@@@���3�?���3�k��8�
>�:���{i��#���Ld�n�<�B$�=6_��ސ��8�þ�Z=��>���;�/�O�>�����q�=VU�����=�l��e,�-X7������xQ>>�K��$>46���a><9�>�(>�(O>�(>�X�<��=�p��D��m��>���(z����x��;W�}J�;~7���:���G���%�=�>	8�<;��ݾ=��U>;H<8B%����Å2���?���=g0ѽ ������$־ڜ^>w����.��c��؟=Tgx�C��;�o:�z���)�=������eٳ>�
f�9ˡ=���<
-���)��R��=/)�<��z>��H�$5���>�����}y��l�>��[��@�;�Ӵ=�0�=8׾T;>яV�?h���>Y�k=��=�꽕!�>5�=PO�~���L�J=��̾�м.�<�W?�;��[.��h���?k�Q���b>�u=�;��+?�<IE�>��>wz�=���	��s�^>��O>�_>Wo�>t����f����%V>��c<B��>�1c�Q�,�j�<=���>�̭��G�$��<��>0m¾����j�x�q��>�Y�=��>�^1=��Ͼ<�<ξTP��}:�j�>C�
>��=��$>J��>ߖ>��>ג�>�
g>�>������7=�|=�J?Q�=��> �j>
�P>-�l>೘�W^�]�)>��w>��=���*X��|�>4��?�X>�]�ώ�M�N��>�>B��8�)`��/=�G�=���>�2�>����>�>�_�>���<}�n��f>`�!��(Q�qr1�C�6��.�>��
>}���%�Ͼ�Ɖ>�ܘ�N��ӶG<f1��d�?0�K>�꾀"k��Ϭ>���>^�?��'�+�e(��s��c��>7���J����&����>��e�D�(=]����q�rJ�>.y>�(����eϪ>�(>;�,==5�/�>S%>2}>n#n=�C=�i>y���d��=��ν�|�8[)���0�=L��
-$�J�:>E�>��.��!��4�y=+r�>�|���t��}����~����>*%��+>0L��fB>[g��	����>>��=$ݽ=މ?>�z@�y
v�L��>�2h>Bl#>L(��#vZ>�ر���<���=s%�>jx?���=
p�<3ś=�5^���>��=�D->��)>)�˽g����s?H���ε�xc���Ҿ �M?hRӾ�D�S��P*,?�Ծ���f#�K��mC&?�(@?Pꬾ�ാ�x�>Z�?J��>��پk���s�u��&�����>] ��鸉���"���@?-r��Ҝ�x���F}���5�>�T�>��+�����(.?���>U���4n�=�w��`�*�$5>q6�>O�Z�=��=�k�=�L�<&q= �u>d#
-=M���&����ꖾy:;>��²{>6ƀ<:�h�G^�������T���1�>N��[�s�� =�:!>o{>4���A䛾lo�� �]>{�v�`n�c=4���f?�>M���0
m>�X>����Ҷ�<'�>����NV����|>�ꑾh�ɾ4#�悞<'K�>�
���~�=HO�>N8B��Ui�K	�=u��=e`?��+�ǟ;���.�?�)�DP\?�-���@�;-�9V\?N�<��\2��5�ؗ'���e?��W?bl+�\�/��5[?��]?�O����A?L?^?�;E�'�
?n�>{�8��)	?��
?��?�Ik?��>�F6�OI����.@?)~?��>��8���?�-?�$@�=�>�>?)�4��_�+���9?��?�?��?��M?L?�7�h�?�<?8�@

?�?G;�vI&�`<�?�B?d��U��s�??o	�>CI
-?ݪ5��?��?��?���
�~��>�=?wC�f��L�W?
+vector_observation����

action_masks����

policy_1/concat_2/concatactionaction_output_shape������?
+action_output_shape




memory_size





version_number




trainer_patch_version




trainer_minor_version




trainer_major_version




is_continuous_control




policy_1/add_2/y




policy_1/Sum/reduction_indices




policy_1/add/y



	
$policy/main_graph_0/hidden_0/BiasAdd
�����?
vector_observation#policy/main_graph_0/hidden_0/kernel


+�
!policy/main_graph_0/hidden_0/bias


�
 policy/main_graph_0/hidden_0/Mul2	�����?
$policy/main_graph_0/hidden_0/BiasAddpolicy_1/dense/MatMul
�����?
 policy/main_graph_0/hidden_0/Mulpolicy/dense/kernel

�
<policy_1/dense/MatMul/patch:0


�
policy_1/strided_slice�



�����?
action_maskspolicy_1/Softmax2�����?
policy_1/dense/MatMulpolicy_1/addd�����?policy_1/Softmaxpolicy_1/add/ypolicy_1/Mulf�����?policy_1/addpolicy_1/strided_slicepolicy_1/Sum������?policy_1/Mulpolicy_1/Sum/reduction_indicespolicy_1/truedivg�����?policy_1/Mulpolicy_1/Sumpolicy_1/add_2d�����?policy_1/truedivpolicy_1/add_2/ypolicy_1/Log_12r�����?
policy_1/add_2policy_1/concat_2/concat2�����?
policy_1/Log_1action2�����?
policy_1/concat_2/concat@@@�A���3�?���3��ƽl>_jX=�P��
+��z�>����0>�����K�>s�>��Y>&�j>�M�����=,�����"����>�P�����>�ɹ<����8��c�==�����:)�y�x�>|\-<���xdO���"����{?=���=[�1>�l>>_����->�8���8��
ž=�m>M�/>����H�=/��Nh���b_�4��>�]�܌<��
>�Ӥ>�E��D>�T=�zO�����A�;Qd���Ͼ�{|��D;>Z�`�/�aV���o�>TJ�=Sa��xİ�5!)>�8�=� 侥Z�='�.�O�<��5n�
A��|�=�����#��jfs>1x=��<)���g����K�>>z>���>���=�< Lb���>�ݘ>X=C=^JѾ��վ*�>�L�>�����`��%T�(Qz��'�>��k�����C�|>0�D�S:3<̋�>�D^=K8�Y^�>kX����f=b�=�׍�J9�=������=t:��
+q�>�h`� �>�J��F��4.��
��[@�;��+>}=��m>�В<xxW��ᔾ+�Q���>��=���>���>E�=��j�>�a�<����Ə<���u���,��FF��VG���
T�m^���&<�+g=Z	^>��u��>tFy>�f�=/{M>����վv��x���n)�=9�A>s�J���@>*�	��:�>��>A��P���q�>qI=�Z���H���l���e>��[>O�þ��:�G����(�)��>���>����G�����>��^=
+�?�ٻ&��B�$>AW5>�٧���?s�=�E�>č�>�-/<��¾�{��#!O��u0>z�<>���>��=+�
�r|�kV�fo�>��=F@Z>�j=V	�>
�6b��U�6='�?sզ��2�;M4��@�C>��?�-
?	�
=x/>��ؾ�g5=c��=�v����c>$Bi>m�����I>ؖ��ou<=�o��cb����ؽ�m�EQ�S�K=�7�=�[��Q�>/*��7��>�j��#|>D�����|��=���W����	�>���=M�6=�W�>�<�>��=����>(���o��>U)�>���>�����-��2�>F>�<ؾ[�7?*��>��E>>1?i��>S�>�# ?�?���>uTJ�W#�>+Y澷&)?�D?��L?���x���d��>��>:'�>�u?�?r_�>�9@?j����>��>���Z
�=��7��Ϟ�@�d�v>�>9O�>��z>H���ᝡ���p=��=J��>��?i(�>8���>%>��%��樾9���n����r>6^->�Х=Ng>�r>�ad>,{��
��>1��<1����!>ev�Z�=+-�����A����<��:>�1f�{J���c=��!�	���]3>�]R�+��<��C>DY0�
+Ty=��)��R�>C�>�[�=޽mU�>?��=�(��:��λ�&.{=��f��3/��kK��\�=W�ʻ1��>�����4�=k�)��������=P�#���,���A�sJ,��ՙ>�kY�>����ܼ~I>%�S���ս�k�վW�g>�f>
+ƙ����l��..R>��ʽ�k�=�=����a�P��IO?��I?�C?��F?t�B?;.�CM?o
(��I?o�F?VGD?h�(����SH?�{H?��>��D?�G?�A?E7F?7���"Ǿk0/?W��
ž�-,?X���
k���!?ߥ߾�����,?Ӿ�⾡�?�T?�.�>��"��N�Y�˾5n0?�?�>*��>%��P�����>.?�l
�b�����?����"	��L-?�H�>x?L"�n��>%~�>�g'��X����^�??��<�ݾ��+?LD!<.V<��,������Q"?,�|����,?G��������?���G��W�-?

README.md

 | **Version** | **Release Date** | **Source** | **Documentation** | **Download** |
 |:-------:|:------:|:-------------:|:-------:|:------------:|
 | **master (unstable)** | -- | [source](https://github.com/Unity-Technologies/ml-agents/tree/master) | [docs](https://github.com/Unity-Technologies/ml-agents/tree/master/docs/Readme.md) | [download](https://github.com/Unity-Technologies/ml-agents/archive/master.zip) |
-| **Release 2** | **May 20, 2020** | **[source](https://github.com/Unity-Technologies/ml-agents/tree/release_2)** | **[docs](https://github.com/Unity-Technologies/ml-agents/tree/release_2/docs/Readme.md)** | **[download](https://github.com/Unity-Technologies/ml-agents/archive/release_2.zip)** |
-| **Release 1** | April 30, 2020 | [source](https://github.com/Unity-Technologies/ml-agents/tree/release_1) | [docs](https://github.com/Unity-Technologies/ml-agents/tree/release_1/docs/Readme.md) | [download](https://github.com/Unity-Technologies/ml-agents/archive/release_1.zip) |
+| **Release 3** | **June 10, 2020** | **[source](https://github.com/Unity-Technologies/ml-agents/tree/release_3)** | **[docs](https://github.com/Unity-Technologies/ml-agents/tree/release_3_docs/docs/Readme.md)** | **[download](https://github.com/Unity-Technologies/ml-agents/archive/release_3.zip)** |
+| **Release 2** | May 20, 2020 | [source](https://github.com/Unity-Technologies/ml-agents/tree/release_2) | [docs](https://github.com/Unity-Technologies/ml-agents/tree/release_2_docs/docs/Readme.md) | [download](https://github.com/Unity-Technologies/ml-agents/archive/release_2.zip) |
+| **Release 1** | April 30, 2020 | [source](https://github.com/Unity-Technologies/ml-agents/tree/release_1) | [docs](https://github.com/Unity-Technologies/ml-agents/tree/release_1_docs/docs/Readme.md) | [download](https://github.com/Unity-Technologies/ml-agents/archive/release_1.zip) |
 | **0.15.1** | March 30, 2020 | [source](https://github.com/Unity-Technologies/ml-agents/tree/0.15.1) | [docs](https://github.com/Unity-Technologies/ml-agents/tree/0.15.1/docs/Readme.md) | [download](https://github.com/Unity-Technologies/ml-agents/archive/0.15.1.zip) |
 | **0.15.0** | March 18, 2020 | [source](https://github.com/Unity-Technologies/ml-agents/tree/0.15.0) | [docs](https://github.com/Unity-Technologies/ml-agents/tree/0.15.0/docs/Readme.md) | [download](https://github.com/Unity-Technologies/ml-agents/archive/0.15.0.zip) |
 | **0.14.1** | February 26, 2020 | [source](https://github.com/Unity-Technologies/ml-agents/tree/0.14.1) | [docs](https://github.com/Unity-Technologies/ml-agents/tree/0.14.1/docs/Readme.md) | [download](https://github.com/Unity-Technologies/ml-agents/archive/0.14.1.zip) |

com.unity.ml-agents/CHANGELOG.md

 - `max_step` in the `TerminalStep` and `TerminalSteps` objects was renamed `interrupted`.
 - `beta` and `epsilon` in `PPO` are no longer decayed by default but follow the same schedule as learning rate. (#3940)
 - `get_behavior_names()` and `get_behavior_spec()` on UnityEnvironment were replaced by the `behavior_specs` property. (#3946)
-- The first version of the Unity Environment Registry (Experimental) has been released. More information [here](https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Unity-Environment-Registry.md)(#3967)
+- The first version of the Unity Environment Registry (Experimental) has been released. More information [here](https://github.com/Unity-Technologies/ml-agents/blob/release_3_docs/docs/Unity-Environment-Registry.md)(#3967)
 - `use_visual` and `allow_multiple_visual_obs` in the `UnityToGymWrapper` constructor
 were replaced by `allow_multiple_obs` which allows one or more visual observations and
 vector observations to be used simultaneously. (#3981) Thank you @shakenes !
 - The format for trainer configuration has changed, and the "default" behavior has been deprecated.
-  See the [Migration Guide](https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Migrating.md) for more details. (#3936)
+  See the [Migration Guide](https://github.com/Unity-Technologies/ml-agents/blob/release_3_docs/docs/Migrating.md) for more details. (#3936)
 - Training artifacts (trained models, summaries) are now found in the `results/`
   directory. (#3829)
 - When using Curriculum, the current lesson will resume if training is quit and resumed. As such,
   - Introduced the `SideChannelManager` to register, unregister and access side
     channels. (#3807)
   - `Academy.FloatProperties` was replaced by `Academy.EnvironmentParameters`.
-    See the [Migration Guide](https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Migrating.md)
+    See the [Migration Guide](https://github.com/Unity-Technologies/ml-agents/blob/release_1_docs/docs/Migrating.md)
     for more details on upgrading. (#3807)
   - `SideChannel.OnMessageReceived` is now a protected method (was public)
   - SideChannel IncomingMessages methods now take an optional default argument,
   `--load`. (#3705)
 - The Jupyter notebooks have been removed from the repository. (#3704)
 - The multi-agent gym option was removed from the gym wrapper. For multi-agent
-  scenarios, use the [Low Level Python API](https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Python-API.md). (#3681)
+  scenarios, use the [Low Level Python API](https://github.com/Unity-Technologies/ml-agents/blob/release_1_docs/docs/Python-API.md). (#3681)
-  [Low Level Python API](https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Python-API.md)
+  [Low Level Python API](https://github.com/Unity-Technologies/ml-agents/blob/release_1_docs/docs/Python-API.md)
   documentation for more information. If you use `mlagents-learn` for training, this should be a
   transparent change. (#3681)
 - Added ability to start training (initialize model weights) from a previous run

com.unity.ml-agents/Runtime/Academy.cs

                         port = port
                     }
                 );
-                Communicator.QuitCommandReceived += OnQuitCommandReceived;
-                Communicator.ResetCommandReceived += OnResetCommand;
             }
 
             if (Communicator != null)
                     );
                     Communicator = null;
                 }
+            }
+            if (Communicator != null)
+            {
+                Communicator.QuitCommandReceived += OnQuitCommandReceived;
+                Communicator.ResetCommandReceived += OnResetCommand;
             }
 
             // If a communicator is enabled/provided, then we assume we are in

com.unity.ml-agents/Runtime/Agent.cs

         /// <seealso cref="OnActionReceived(float[])"/> function, which receives this array and
         /// implements the corresponding agent behavior. See [Actions] for more information
         /// about agent actions.
+        /// Note : Do not create a new float array of action in the `Heuristic()` method,
+        /// as this will prevent writing floats to the original action array.
         ///
         /// An agent calls this `Heuristic()` function to make a decision when you set its behavior
         /// type to <see cref="BehaviorType.HeuristicOnly"/>. The agent also calls this function if

config/imitation/CrawlerStatic.yaml

         use_actions: false
         use_vail: false
         demo_path: Project/Assets/ML-Agents/Examples/Crawler/Demos/ExpertCrawlerSta.demo
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 1000

config/imitation/FoodCollector.yaml

         use_actions: false
         use_vail: false
         demo_path: Project/Assets/ML-Agents/Examples/FoodCollector/Demos/ExpertFood.demo
-    output_path: default
     keep_checkpoints: 5
     max_steps: 2000000
     time_horizon: 64

config/imitation/Hallway.yaml

         use_actions: false
         use_vail: false
         demo_path: Project/Assets/ML-Agents/Examples/Hallway/Demos/ExpertHallway.demo
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 64

config/imitation/PushBlock.yaml

         use_actions: false
         use_vail: false
         demo_path: Project/Assets/ML-Agents/Examples/PushBlock/Demos/ExpertPush.demo
-    output_path: default
     keep_checkpoints: 5
     max_steps: 15000000
     time_horizon: 64

config/ppo/3DBall.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 500000
     time_horizon: 1000

config/ppo/3DBallHard.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 5000000
     time_horizon: 1000

config/ppo/3DBall_randomize.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 500000
     time_horizon: 1000

config/ppo/Basic.yaml

       extrinsic:
         gamma: 0.9
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 500000
     time_horizon: 3

config/ppo/Bouncer.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 4000000
     time_horizon: 64

config/ppo/CrawlerDynamic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 1000

config/ppo/CrawlerStatic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 1000

config/ppo/FoodCollector.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 2000000
     time_horizon: 64

config/ppo/GridWorld.yaml

       extrinsic:
         gamma: 0.9
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 500000
     time_horizon: 5

config/ppo/Hallway.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 64

config/ppo/PushBlock.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 2000000
     time_horizon: 64

config/ppo/Pyramids.yaml

         strength: 0.02
         encoding_size: 256
         learning_rate: 0.0003
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 128

config/ppo/Reacher.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 20000000
     time_horizon: 1000

config/ppo/SoccerTwos.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 50000000
     time_horizon: 1000

config/ppo/StrikersVsGoalie.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 50000000
     time_horizon: 1000
       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 50000000
     time_horizon: 1000

config/ppo/Tennis.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 50000000
     time_horizon: 1000

config/ppo/VisualHallway.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 64

config/ppo/VisualPushBlock.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 3000000
     time_horizon: 64

config/ppo/VisualPyramids.yaml

         strength: 0.01
         encoding_size: 256
         learning_rate: 0.0003
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 128

config/ppo/WalkerDynamic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 25000000
     time_horizon: 1000

config/ppo/WalkerStatic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 20000000
     time_horizon: 1000

config/ppo/WallJump.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 20000000
     time_horizon: 128
       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 5000000
     time_horizon: 128

config/ppo/WallJump_curriculum.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 20000000
     time_horizon: 128
       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 5000000
     time_horizon: 128

config/ppo/WormDynamic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 3500000
     time_horizon: 1000

config/ppo/WormStatic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 3500000
     time_horizon: 1000

config/sac/3DBall.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 500000
     time_horizon: 1000

config/sac/3DBallHard.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 500000
     time_horizon: 1000

config/sac/Basic.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 500000
     time_horizon: 10

config/sac/Bouncer.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 1000000
     time_horizon: 64

config/sac/CrawlerDynamic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 5000000
     time_horizon: 1000

config/sac/CrawlerStatic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 3000000
     time_horizon: 1000

config/sac/FoodCollector.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 2000000
     time_horizon: 64

config/sac/GridWorld.yaml

       extrinsic:
         gamma: 0.9
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 500000
     time_horizon: 5

config/sac/Hallway.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 5000000
     time_horizon: 64

config/sac/PushBlock.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 2000000
     time_horizon: 64

config/sac/Pyramids.yaml

         use_actions: true
         use_vail: false
         demo_path: Project/Assets/ML-Agents/Examples/Pyramids/Demos/ExpertPyramid.demo
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 128

config/sac/Reacher.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 20000000
     time_horizon: 1000

config/sac/Tennis.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 20000000
     time_horizon: 64

config/sac/VisualHallway.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 64

config/sac/VisualPushBlock.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 3000000
     time_horizon: 64

config/sac/VisualPyramids.yaml

         use_actions: true
         use_vail: false
         demo_path: Project/Assets/ML-Agents/Examples/Pyramids/Demos/ExpertPyramid.demo
-    output_path: default
     keep_checkpoints: 5
     max_steps: 10000000
     time_horizon: 128

config/sac/WalkerDynamic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 15000000
     time_horizon: 1000

config/sac/WalkerStatic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 20000000
     time_horizon: 1000

config/sac/WallJump.yaml

       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 15000000
     time_horizon: 128
       extrinsic:
         gamma: 0.99
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 5000000
     time_horizon: 128

config/sac/WormDynamic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 5000000
     time_horizon: 1000

config/sac/WormStatic.yaml

       extrinsic:
         gamma: 0.995
         strength: 1.0
-    output_path: default
     keep_checkpoints: 5
     max_steps: 3000000
     time_horizon: 1000

docs/Learning-Environment-Design-Agents.md

 - `Agent.Heuristic()` - When the `Behavior Type` is set to `Heuristic Only` in
   the Behavior Parameters of the Agent, the Agent will use the `Heuristic()`
   method to generate the actions of the Agent. As such, the `Heuristic()` method
-  writes to a provided array of floats.
+  writes to the array of floats provided to the Heuristic method as argument.
+  __Note__: Do not create a new float array of action in the `Heuristic()` method,
+  as this will prevent writing floats to the original action array.
 
 As a concrete example, here is how the Ball3DAgent class implements these methods:
 

ml-agents/mlagents/trainers/ghost/trainer.py

         reward_buff_cap,
         trainer_settings,
         training,
-        run_id,
+        artifact_path,
     ):
         """
         Creates a GhostTrainer.
         :param reward_buff_cap: Max reward history to track in the reward buffer
         :param trainer_settings: The parameters for the trainer.
         :param training: Whether the trainer is set for training.
-        :param run_id: The identifier of the current run
+        :param artifact_path: Path to store artifacts from this trainer.
-            brain_name, trainer_settings, training, run_id, reward_buff_cap
+            brain_name, trainer_settings, training, artifact_path, reward_buff_cap
         )
 
         self.trainer = trainer

ml-agents/mlagents/trainers/learn.py

         )
         trainer_factory = TrainerFactory(
             options.behaviors,
-            checkpoint_settings.run_id,
             write_path,
             not checkpoint_settings.inference,
             checkpoint_settings.resume,

ml-agents/mlagents/trainers/policy/nn_policy.py

         brain: BrainParameters,
         trainer_params: TrainerSettings,
         is_training: bool,
+        model_path: str,
         load: bool,
         tanh_squash: bool = False,
         reparameterize: bool = False,
         :param trainer_params: Defined training parameters.
         :param is_training: Whether the model should be trained.
         :param load: Whether a pre-trained model will be loaded or a new one created.
+        :param model_path: Path where the model should be saved and loaded.
-        super().__init__(seed, brain, trainer_params, load)
+        super().__init__(seed, brain, trainer_params, model_path, load)
         self.grads = None
         self.update_batch: Optional[tf.Operation] = None
         num_layers = self.network_settings.num_layers

ml-agents/mlagents/trainers/policy/tf_policy.py

         seed: int,
         brain: BrainParameters,
         trainer_settings: TrainerSettings,
+        model_path: str,
         load: bool = False,
     ):
         """
         :param trainer_settings: The trainer parameters.
+        :param model_path: Where to load/save the model.
+        :param load: If True, load model from model_path. Otherwise, create new model.
         """
 
         self.m_size = 0
         self.use_continuous_act = brain.vector_action_space_type == "continuous"
         if self.use_continuous_act:
             self.num_branches = self.brain.vector_action_space_size[0]
-        self.model_path = self.trainer_settings.output_path
+        self.model_path = model_path
         self.initialize_path = self.trainer_settings.init_path
         self.keep_checkpoints = self.trainer_settings.keep_checkpoints
         self.graph = tf.Graph()

ml-agents/mlagents/trainers/ppo/trainer.py

         training: bool,
         load: bool,
         seed: int,
-        run_id: str,
+        artifact_path: str,
     ):
         """
         Responsible for collecting experiences and training PPO model.
         :param training: Whether the trainer is set for training.
         :param load: Whether the model should be loaded.
         :param seed: The seed the model will be initialized with
-        :param run_id: The identifier of the current run
+        :param artifact_path: The directory within which to store artifacts from this trainer.
-            brain_name, trainer_settings, training, run_id, reward_buff_cap
+            brain_name, trainer_settings, training, artifact_path, reward_buff_cap
         )
         self.hyperparameters: PPOSettings = cast(
             PPOSettings, self.trainer_settings.hyperparameters
             brain_parameters,
             self.trainer_settings,
             self.is_training,
+            self.artifact_path,
             self.load,
             condition_sigma_on_obs=False,  # Faster training for PPO
             create_tf_graph=False,  # We will create the TF graph in the Optimizer

ml-agents/mlagents/trainers/sac/trainer.py

         training: bool,
         load: bool,
         seed: int,
-        run_id: str,
+        artifact_path: str,
     ):
         """
         Responsible for collecting experiences and training SAC model.
         :param training: Whether the trainer is set for training.
         :param load: Whether the model should be loaded.
         :param seed: The seed the model will be initialized with
-        :param run_id: The identifier of the current run
+        :param artifact_path: The directory within which to store artifacts from this trainer.
-            brain_name, trainer_settings, training, run_id, reward_buff_cap
+            brain_name, trainer_settings, training, artifact_path, reward_buff_cap
         )
 
         self.load = load
         """
         Save the training buffer's update buffer to a pickle file.
         """
-        filename = os.path.join(
-            self.trainer_settings.output_path, "last_replay_buffer.hdf5"
-        )
+        filename = os.path.join(self.artifact_path, "last_replay_buffer.hdf5")
         logger.info("Saving Experience Replay Buffer to {}".format(filename))
         with open(filename, "wb") as file_object:
             self.update_buffer.save_to_file(file_object)
         Loads the last saved replay buffer from a file.
         """
-        filename = os.path.join(
-            self.trainer_settings.output_path, "last_replay_buffer.hdf5"
-        )
+        filename = os.path.join(self.artifact_path, "last_replay_buffer.hdf5")
         logger.info("Loading Experience Replay Buffer from {}".format(filename))
         with open(filename, "rb+") as file_object:
             self.update_buffer.load_from_file(file_object)
             brain_parameters,
             self.trainer_settings,
             self.is_training,
+            self.artifact_path,
             self.load,
             tanh_squash=True,
             reparameterize=True,

ml-agents/mlagents/trainers/settings.py

         factory=lambda: {RewardSignalType.EXTRINSIC: RewardSignalSettings()}
     )
     init_path: Optional[str] = None
-    output_path: str = "default"
     keep_checkpoints: int = 5
     checkpoint_interval: int = 500000
     max_steps: int = 500000
         REWARD: str = "reward"
 
     measure: str = attr.ib(default=MeasureType.REWARD)
-    thresholds: List[int] = attr.ib(factory=list)
+    thresholds: List[float] = attr.ib(factory=list)
     min_lesson_length: int = 0
     signal_smoothing: bool = True
     parameters: Dict[str, List[float]] = attr.ib(kw_only=True)

ml-agents/mlagents/trainers/tests/test_barracuda_converter.py

 @pytest.mark.parametrize("rnn", [True, False], ids=["rnn", "no_rnn"])
 def test_policy_conversion(tmpdir, rnn, visual, discrete):
     tf.reset_default_graph()
-    dummy_config = TrainerSettings(output_path=os.path.join(tmpdir, "test"))
+    dummy_config = TrainerSettings()
-        dummy_config, use_rnn=rnn, use_discrete=discrete, use_visual=visual
+        dummy_config,
+        use_rnn=rnn,
+        model_path=os.path.join(tmpdir, "test"),
+        use_discrete=discrete,
+        use_visual=visual,
     )
     policy.save_model(1000)
     settings = SerializationSettings(

ml-agents/mlagents/trainers/tests/test_bcmodule.py

         NetworkSettings.MemorySettings() if use_rnn else None
     )
     policy = NNPolicy(
-        0, mock_brain, trainer_config, False, False, tanhresample, tanhresample
+        0, mock_brain, trainer_config, False, "test", False, tanhresample, tanhresample
     )
     with policy.graph.as_default():
         bc_module = BCModule(

ml-agents/mlagents/trainers/tests/test_meta_curriculum.py

 import pytest
 from unittest.mock import patch, Mock, call
+import yaml
+import cattr
 
 from mlagents.trainers.meta_curriculum import MetaCurriculum
 
 @pytest.fixture
 def reward_buff_sizes():
     return {"Brain1": 7, "Brain2": 8}
+
+
+def test_convert_from_dict():
+    config = yaml.safe_load(
+        """
+        measure: progress
+        thresholds: [0.1, 0.3, 0.5]
+        min_lesson_length: 100
+        signal_smoothing: true
+        parameters:
+            param1: [0.0, 4.0, 6.0, 8.0]
+        """
+    )
+    should_be_config = CurriculumSettings(
+        thresholds=[0.1, 0.3, 0.5],
+        min_lesson_length=100,
+        signal_smoothing=True,
+        measure=CurriculumSettings.MeasureType.PROGRESS,
+        parameters={"param1": [0.0, 4.0, 6.0, 8.0]},
+    )
+    assert cattr.structure(config, CurriculumSettings) == should_be_config
 
 
 def test_curriculum_config(param_name="test_param1", min_lesson_length=100):

ml-agents/mlagents/trainers/tests/test_nn_policy.py

     use_rnn: bool = False,
     use_discrete: bool = True,
     use_visual: bool = False,
+    model_path: str = "",
     load: bool = False,
     seed: int = 0,
 ) -> NNPolicy:
     trainer_settings.network_settings.memory = (
         NetworkSettings.MemorySettings() if use_rnn else None
     )
-    policy = NNPolicy(seed, mock_brain, trainer_settings, False, load)
+    policy = NNPolicy(seed, mock_brain, trainer_settings, False, model_path, load)
     return policy
 
 
-    trainer_params = TrainerSettings(output_path=path1)
-    policy = create_policy_mock(trainer_params)
+    trainer_params = TrainerSettings()
+    policy = create_policy_mock(trainer_params, model_path=path1)
     policy.initialize_or_load()
     policy._set_step(2000)
     policy.save_model(2000)
     # Try load from this path
-    policy2 = create_policy_mock(trainer_params, load=True, seed=1)
+    policy2 = create_policy_mock(trainer_params, model_path=path1, load=True, seed=1)
     policy2.initialize_or_load()
     _compare_two_policies(policy, policy2)
     assert policy2.get_current_step() == 2000
     trainer_params.init_path = path1
-    policy3 = create_policy_mock(trainer_params, load=False, seed=2)
+    policy3 = create_policy_mock(trainer_params, model_path=path1, load=False, seed=2)
     policy3.initialize_or_load()
 
     _compare_two_policies(policy2, policy3)
         # Test write_stats
         with self.assertLogs("mlagents.trainers", level="WARNING") as cm:
             path1 = tempfile.mkdtemp()
-            trainer_params = TrainerSettings(output_path=path1)
-            policy = create_policy_mock(trainer_params)
+            trainer_params = TrainerSettings()
+            policy = create_policy_mock(trainer_params, model_path=path1)
             policy.initialize_or_load()
             policy._check_model_version(
                 "0.0.0"
         brain_params,
         TrainerSettings(network_settings=NetworkSettings(normalize=True)),
         False,
+        "testdir",
         False,
     )
 

ml-agents/mlagents/trainers/tests/test_policy.py

 
 def test_take_action_returns_empty_with_no_agents():
     test_seed = 3
-    policy = FakePolicy(test_seed, basic_mock_brain(), TrainerSettings())
+    policy = FakePolicy(test_seed, basic_mock_brain(), TrainerSettings(), "output")
     # Doesn't really matter what this is
     dummy_groupspec = BehaviorSpec([(1,)], "continuous", 1)
     no_agent_step = DecisionSteps.empty(dummy_groupspec)
 
 def test_take_action_returns_nones_on_missing_values():
     test_seed = 3
-    policy = FakePolicy(test_seed, basic_mock_brain(), TrainerSettings())
+    policy = FakePolicy(test_seed, basic_mock_brain(), TrainerSettings(), "output")
     policy.evaluate = MagicMock(return_value={})
     policy.save_memories = MagicMock()
     step_with_agents = DecisionSteps(
 
 def test_take_action_returns_action_info_when_available():
     test_seed = 3
-    policy = FakePolicy(test_seed, basic_mock_brain(), TrainerSettings())
+    policy = FakePolicy(test_seed, basic_mock_brain(), TrainerSettings(), "output")
     policy_eval_out = {
         "action": np.array([1.0], dtype=np.float32),
         "memory_out": np.array([[2.5]], dtype=np.float32),

ml-agents/mlagents/trainers/tests/test_ppo.py

         else None
     )
     policy = NNPolicy(
-        0, mock_brain, trainer_settings, False, False, create_tf_graph=False
+        0, mock_brain, trainer_settings, False, "test", False, create_tf_graph=False
     )
     optimizer = PPOOptimizer(policy, trainer_settings)
     return optimizer

ml-agents/mlagents/trainers/tests/test_reward_signals.py

         else None
     )
     policy = NNPolicy(
-        0, mock_brain, trainer_settings, False, False, create_tf_graph=False
+        0, mock_brain, trainer_settings, False, "test", False, create_tf_graph=False
     )
     if trainer_settings.trainer_type == TrainerType.SAC:
         optimizer = SACOptimizer(policy, trainer_settings)

ml-agents/mlagents/trainers/tests/test_sac.py

         else None
     )
     policy = NNPolicy(
-        0, mock_brain, trainer_settings, False, False, create_tf_graph=False
+        0, mock_brain, trainer_settings, False, "test", False, create_tf_graph=False
     )
     optimizer = SACOptimizer(policy, trainer_settings)
     return optimizer
     )
     trainer_params = dummy_config
     trainer_params.hyperparameters.save_replay_buffer = True
-    trainer = SACTrainer(mock_brain.brain_name, 1, trainer_params, True, False, 0, 0)
+    trainer = SACTrainer(
+        mock_brain.brain_name, 1, trainer_params, True, False, 0, "testdir"
+    )
     policy = trainer.create_policy(mock_brain.brain_name, mock_brain)
     trainer.add_policy(mock_brain.brain_name, policy)
 
 
     # Wipe Trainer and try to load
-    trainer2 = SACTrainer(mock_brain.brain_name, 1, trainer_params, True, True, 0, 0)
+    trainer2 = SACTrainer(
+        mock_brain.brain_name, 1, trainer_params, True, True, 0, "testdir"
+    )
 
     policy = trainer2.create_policy(mock_brain.brain_name, mock_brain)
     trainer2.add_policy(mock_brain.brain_name, policy)

ml-agents/mlagents/trainers/tests/test_simple_rl.py

             env_manager = SimpleEnvManager(env, EnvironmentParametersChannel())
         trainer_factory = TrainerFactory(
             trainer_config=trainer_config,
-            run_id=run_id,
             output_path=dir,
             train_model=True,
             load_model=False,

ml-agents/mlagents/trainers/tests/test_trainer_util.py

     brain_params_mock = BrainParametersMock()
     BrainParametersMock.return_value.brain_name = "testbrain"
     external_brains = {"testbrain": BrainParametersMock()}
-    run_id = "testrun"
     output_path = "results_dir"
     train_model = True
     load_model = False
     expected_config = PPO_CONFIG
 
     def mock_constructor(
-        self, brain, reward_buff_cap, trainer_settings, training, load, seed, run_id
+        self,
+        brain,
+        reward_buff_cap,
+        trainer_settings,
+        training,
+        load,
+        seed,
+        artifact_path,
     ):
         assert brain == brain_params_mock.brain_name
         assert trainer_settings == expected_config
         assert seed == seed
-        assert run_id == run_id
+        assert artifact_path == os.path.join(output_path, brain_name)
-            run_id=run_id,
             output_path=output_path,
             train_model=train_model,
             load_model=load_model,
 
     trainer_factory = trainer_util.TrainerFactory(
         trainer_config=no_default_config,
-        run_id="testrun",
         output_path="output_path",
         train_model=True,
         load_model=False,

ml-agents/mlagents/trainers/trainer/trainer.py

         brain_name: str,
         trainer_settings: TrainerSettings,
         training: bool,
-        run_id: str,
+        artifact_path: str,
-        :dict trainer_settings: The parameters for the trainer (dictionary).
-        :bool training: Whether the trainer is set for training.
-        :str run_id: The identifier of the current run
-        :int reward_buff_cap:
+        :param trainer_settings: The parameters for the trainer (dictionary).
+        :param training: Whether the trainer is set for training.
+        :param artifact_path: The directory within which to store artifacts from this trainer
+        :param reward_buff_cap:
-        self.run_id = run_id
         self.trainer_settings = trainer_settings
         self._threaded = trainer_settings.threaded
         self._stats_reporter = StatsReporter(brain_name)
         self.trajectory_queues: List[AgentManagerQueue[Trajectory]] = []
         self.step: int = 0
+        self.artifact_path = artifact_path
         self.summary_freq = self.trainer_settings.summary_freq
 
     @property

ml-agents/mlagents/trainers/trainer_util.py

     def __init__(
         self,
         trainer_config: Dict[str, TrainerSettings],
-        run_id: str,
         output_path: str,
         train_model: bool,
         load_model: bool,
         multi_gpu: bool = False,
     ):
         self.trainer_config = trainer_config
-        self.run_id = run_id
         self.output_path = output_path
         self.init_path = init_path
         self.train_model = train_model
         return initialize_trainer(
             self.trainer_config[brain_name],
             brain_name,
-            self.run_id,
             self.output_path,
             self.train_model,
             self.load_model,
 def initialize_trainer(