Move gym from habitat_baselines to habitat (facebookresearch#864)

* merging

* -

* Updating configs

* -

* -

* -

* fixing tests

* fixing tests

* changing configs

* Update habitat/tasks/rearrange/multi_task/composite_sensors.py

Co-authored-by: Aaron Gokaslan <skylion.aaron@gmail.com>

* Update habitat/tasks/rearrange/multi_task/composite_sensors.py

Co-authored-by: Aaron Gokaslan <skylion.aaron@gmail.com>

* addressing comments

Co-authored-by: vincentpierre <vincentpierre@users.noreply.github.com>
Co-authored-by: Aaron Gokaslan <skylion.aaron@gmail.com>

configs/tasks/rearrange/close_cab.yaml

@@ -1,3 +1,7 @@
+GYM:
+    AUTO_NAME: "CloseCab"
+    OBS_KEYS: ['joint', 'ee_pos', 'is_holding', 'marker_rel_pos', 'marker_js', 'relative_resting_position']
+    CLASS_NAME: "RearrangeRLEnv"
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 200
 DATASET:
@@ -23,9 +27,12 @@ TASK:
     USE_MARKER_T: True
     SUCCESS_STATE: 0.0
 
-    # Measurements for composite tasks.
-    REWARD_MEASUREMENT: "art_obj_reward"
-    SUCCESS_MEASUREMENT: "art_obj_success"
+    REWARD_MEASURE: "art_obj_reward"
+    SUCCESS_MEASURE: "art_obj_success"
+    SUCCESS_REWARD: 10.0
+    SLACK_REWARD: 0.0
+    END_ON_SUCCESS: True
+
 
     # If true, does not care about navigability or collisions with objects when spawning
     # robot

configs/tasks/rearrange/close_fridge.yaml

@@ -1,3 +1,7 @@
+GYM:
+    AUTO_NAME: "CloseFridge"
+    OBS_KEYS: ['joint', 'ee_pos', 'is_holding', 'marker_rel_pos', 'marker_js', 'relative_resting_position']
+    CLASS_NAME: "RearrangeRLEnv"
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 200
 DATASET:
@@ -22,9 +26,11 @@ TASK:
     USE_MARKER_T: False
     SUCCESS_STATE: 0.0
 
-    # Measurements for composite tasks.
-    REWARD_MEASUREMENT: "art_obj_reward"
-    SUCCESS_MEASUREMENT: "art_obj_success"
+    REWARD_MEASURE: "art_obj_reward"
+    SUCCESS_MEASURE: "art_obj_success"
+    SUCCESS_REWARD: 10.0
+    SLACK_REWARD: 0.0
+    END_ON_SUCCESS: True
 
     # If true, does not care about navigability or collisions with objects when spawning
     # robot

configs/tasks/rearrange/nav_pick.yaml

@@ -1,3 +1,7 @@
+GYM:
+    AUTO_NAME: "NavPick"
+    OBS_KEYS: ['robot_head_depth', 'obj_start_sensor', 'joint', 'is_holding', 'relative_resting_position']
+    CLASS_NAME: "RearrangeRLEnv"
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 500
 DATASET:
@@ -7,6 +11,12 @@ DATASET:
     SCENES_DIR: data/replica_cad/
 TASK:
     TYPE: RearrangeCompositeTask-v0
+
+    REWARD_MEASURE: "move_obj_reward"
+    SUCCESS_MEASURE: "composite_success"
+    SUCCESS_REWARD: 100.0
+    SLACK_REWARD: -0.01
+
     MAX_COLLISIONS: -1.0
     COUNT_OBJ_COLLISIONS: true
     COUNT_ROBOT_OBJ_COLLS: false

configs/tasks/rearrange/nav_pick_nav_place.yaml

@@ -1,3 +1,7 @@
+GYM:
+    CLASS_NAME: "RearrangeRLEnv"
+    AUTO_NAME: "NavPickNavPlace"
+    OBS_KEYS: ['robot_head_depth', 'obj_start_sensor', 'joint', 'is_holding', 'relative_resting_position']
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 500
 DATASET:
@@ -7,6 +11,12 @@ DATASET:
     SCENES_DIR: data/replica_cad/
 TASK:
     TYPE: RearrangeCompositeTask-v0
+
+    REWARD_MEASURE: "move_obj_reward"
+    SUCCESS_MEASURE: "composite_success"
+    SUCCESS_REWARD: 100.0
+    SLACK_REWARD: -0.01
+
     MAX_COLLISIONS: -1.0
     COUNT_OBJ_COLLISIONS: true
     COUNT_ROBOT_OBJ_COLLS: false

configs/tasks/rearrange/nav_to_obj.yaml

@@ -1,3 +1,7 @@
+GYM:
+    AUTO_NAME: "NavToObj"
+    OBS_KEYS: ["object_to_agent_gps_compass", 'joint']
+    CLASS_NAME: "RearrangeRLEnv"
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 400
 DATASET:
@@ -27,9 +31,11 @@ TASK:
     OBJ_SUCC_THRESH: 0.3
     ART_SUCC_THRESH: 0.15
 
-    # Measurements for composite tasks.
-    REWARD_MEASUREMENT: "nav_to_obj_reward"
-    SUCCESS_MEASUREMENT: "nav_to_obj_success"
+    REWARD_MEASURE: "nav_to_obj_reward"
+    SUCCESS_MEASURE: "nav_to_obj_success"
+    SUCCESS_REWARD: 10.0
+    SLACK_REWARD: 0.0
+    END_ON_SUCCESS: True
 
     TARGET_START_POINT_GOAL_SENSOR:
         TYPE: "TargetOrGoalStartPointGoalSensor"

configs/tasks/rearrange/open_cab.yaml

@@ -1,3 +1,7 @@
+GYM:
+    AUTO_NAME: "OpenCab"
+    OBS_KEYS: ['joint', 'ee_pos', 'is_holding', 'marker_rel_pos', 'marker_js', 'relative_resting_position']
+    CLASS_NAME: "RearrangeRLEnv"
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 200
 DATASET:
@@ -25,9 +29,11 @@ TASK:
     USE_MARKER_T: True
     SUCCESS_STATE: 0.45
 
-    # Measurements for composite tasks.
-    REWARD_MEASUREMENT: "art_obj_reward"
-    SUCCESS_MEASUREMENT: "art_obj_success"
+    REWARD_MEASURE: "art_obj_reward"
+    SUCCESS_MEASURE: "art_obj_success"
+    SUCCESS_REWARD: 10.0
+    SLACK_REWARD: 0.0
+    END_ON_SUCCESS: True
 
     # If true, does not care about navigability or collisions with objects when spawning
     # robot

configs/tasks/rearrange/open_fridge.yaml

@@ -1,3 +1,7 @@
+GYM:
+    AUTO_NAME: "OpenFridge"
+    OBS_KEYS: ['joint', 'ee_pos', 'is_holding', 'marker_rel_pos', 'marker_js', 'relative_resting_position']
+    CLASS_NAME: "RearrangeRLEnv"
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 200
 DATASET:
@@ -25,9 +29,11 @@ TASK:
     # (pi/2) - 0.35
     SUCCESS_STATE: 1.2207963268
 
-    # Measurements for composite tasks.
-    REWARD_MEASUREMENT: "art_obj_reward"
-    SUCCESS_MEASUREMENT: "art_obj_success"
+    REWARD_MEASURE: "art_obj_reward"
+    SUCCESS_MEASURE: "art_obj_success"
+    SUCCESS_REWARD: 10.0
+    SLACK_REWARD: 0.0
+    END_ON_SUCCESS: True
 
     # If true, does not care about navigability or collisions with objects when spawning
     # robot

configs/tasks/rearrange/pick.yaml

@@ -1,3 +1,7 @@
+GYM:
+    AUTO_NAME: "Pick"
+    OBS_KEYS: ['robot_head_depth', 'obj_start_sensor', 'joint', 'is_holding', 'relative_resting_position']
+    CLASS_NAME: "RearrangeRLEnv"
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 200
 DATASET:
@@ -20,9 +24,11 @@ TASK:
     CONSTRAINT_VIOLATION_ENDS_EPISODE: True
     FORCE_REGENERATE: False
 
-    # Measurements for composite tasks.
-    REWARD_MEASUREMENT: "rearrangepick_reward"
-    SUCCESS_MEASUREMENT: "rearrangepick_success"
+    REWARD_MEASURE: "rearrangepick_reward"
+    SUCCESS_MEASURE: "rearrangepick_success"
+    SUCCESS_REWARD: 100.0
+    SLACK_REWARD: -0.01
+    END_ON_SUCCESS: True
 
     # If true, does not care about navigability or collisions with objects when spawning
     # robot

configs/tasks/rearrange/place.yaml

@@ -1,3 +1,7 @@
+GYM:
+    AUTO_NAME: "Place"
+    OBS_KEYS: ['obj_goal_sensor', 'joint', 'is_holding', 'relative_resting_position']
+    CLASS_NAME: "RearrangeRLEnv"
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 200
 DATASET:
@@ -21,9 +25,10 @@ TASK:
     DESIRED_RESTING_POSITION: [0.5, 0.0, 1.0]
     FORCE_REGENERATE: False
 
-    # Measurements for composite tasks.
-    REWARD_MEASUREMENT: "place_reward"
-    SUCCESS_MEASUREMENT: "place_success"
+    REWARD_MEASURE: "place_reward"
+    SUCCESS_MEASURE: "place_success"
+    SUCCESS_REWARD: 10.0
+    SLACK_REWARD: 0.0
 
     # If true, does not care about navigability or collisions with objects when spawning
     # robot

configs/tasks/rearrange/prepare_groceries.yaml

@@ -1,3 +1,7 @@
+GYM:
+    CLASS_NAME: "RearrangeRLEnv"
+    AUTO_NAME: "PrepareGroceries"
+    OBS_KEYS: ['robot_head_depth', 'obj_start_sensor', 'joint', 'is_holding', 'relative_resting_position']
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 4000
 DATASET:
@@ -7,6 +11,12 @@ DATASET:
     SCENES_DIR: data/replica_cad/
 TASK:
     TYPE: RearrangeCompositeTask-v0
+
+    REWARD_MEASURE: "move_obj_reward"
+    SUCCESS_MEASURE: "composite_success"
+    SUCCESS_REWARD: 100.0
+    SLACK_REWARD: -0.01
+
     MAX_COLLISIONS: -1.0
     COUNT_OBJ_COLLISIONS: true
     COUNT_ROBOT_OBJ_COLLS: false

configs/tasks/rearrange/reach_state.yaml

@@ -1,3 +1,8 @@
+GYM:
+    AUTO_NAME: "ReachState"
+    OBS_KEYS: ['joint', 'relative_resting_position']
+    ACTION_KEYS: ['ARM_ACTION']
+    CLASS_NAME: "RearrangeRLEnv"
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 20
 DATASET:
@@ -13,6 +18,10 @@ TASK:
     EE_SAMPLE_FACTOR: 0.2
     EE_EXCLUDE_REGION: 0.0
 
+    REWARD_MEASURE: "rearrange_reach_reward"
+    SUCCESS_MEASURE: "rearrange_reach_success"
+    SUCCESS_REWARD: 10.0
+
     DESIRED_RESTING_POSITION: [0.5, 0.0, 1.0]
 
     JOINT_SENSOR:

configs/tasks/rearrange/set_table.yaml

@@ -1,3 +1,8 @@
+GYM:
+    AUTO_NAME: "SetTable"
+    CLASS_NAME: "RearrangeRLEnv"
+    OBS_KEYS: ['robot_head_depth', 'obj_start_sensor', 'joint', 'is_holding', 'relative_resting_position']
+
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 4500
 DATASET:
@@ -7,6 +12,12 @@ DATASET:
     SCENES_DIR: data/replica_cad/
 TASK:
     TYPE: RearrangeCompositeTask-v0
+
+    REWARD_MEASURE: "move_obj_reward"
+    SUCCESS_MEASURE: "composite_success"
+    SUCCESS_REWARD: 100.0
+    SLACK_REWARD: -0.01
+
     MAX_COLLISIONS: -1.0
     COUNT_OBJ_COLLISIONS: true
     COUNT_ROBOT_OBJ_COLLS: false

configs/tasks/rearrange/tidy_house.yaml

@@ -1,3 +1,7 @@
+GYM:
+    CLASS_NAME: "RearrangeRLEnv"
+    AUTO_NAME: "TidyHouse"
+    OBS_KEYS: ['robot_head_depth', 'obj_start_sensor', 'joint', 'is_holding', 'relative_resting_position']
 ENVIRONMENT:
     MAX_EPISODE_STEPS: 5000
 DATASET:
@@ -7,6 +11,12 @@ DATASET:
     SCENES_DIR: data/replica_cad/
 TASK:
     TYPE: RearrangeCompositeTask-v0
+
+    REWARD_MEASURE: "move_obj_reward"
+    SUCCESS_MEASURE: "composite_success"
+    SUCCESS_REWARD: 100.0
+    SLACK_REWARD: -0.01
+
     MAX_COLLISIONS: -1.0
     COUNT_OBJ_COLLISIONS: true
     COUNT_ROBOT_OBJ_COLLS: false

examples/interactive_play.py

@@ -54,8 +54,8 @@
 import habitat.tasks.rearrange.rearrange_task
 from habitat.tasks.rearrange.actions import ArmEEAction
 from habitat.tasks.rearrange.utils import euler_to_quat
+from habitat.utils.render_wrapper import overlay_frame
 from habitat.utils.visualizations.utils import observations_to_image
-from habitat_baselines.utils.render_wrapper import overlay_frame
 from habitat_sim.utils import viz_utils as vut
 
 try:

examples/tutorials/colabs/Habitat2_Quickstart.ipynb

@@ -74,14 +74,14 @@
     "import numpy as np\n",
     "\n",
     "import habitat\n",
-    "import habitat_baselines.utils.gym_definitions as habitat_gym\n",
+    "import habitat.utils.gym_definitions as habitat_gym\n",
     "from habitat.core.embodied_task import Measure\n",
     "from habitat.core.registry import registry\n",
     "from habitat.core.simulator import Sensor, SensorTypes\n",
     "from habitat.tasks.rearrange.rearrange_sensors import RearrangeReward\n",
     "from habitat.tasks.rearrange.rearrange_task import RearrangeTask\n",
+    "from habitat.utils.render_wrapper import overlay_frame\n",
     "from habitat.utils.visualizations.utils import observations_to_image\n",
-    "from habitat_baselines.utils.render_wrapper import overlay_frame\n",
     "from habitat_sim.utils import viz_utils as vut\n",
     "\n",
     "# Quiet the Habitat simulator logging\n",

examples/tutorials/colabs/habitat2_gym_tutorial.ipynb

@@ -67,7 +67,7 @@
     "import gym\n",
     "\n",
     "# flake8: noqa\n",
-    "import habitat_baselines.utils.gym_definitions"
+    "import habitat.utils.gym_definitions"
    ]
   },
   {
@@ -168,7 +168,7 @@
     "env = gym.make(\n",
     "    \"HabitatPick-v0\",\n",
     "    override_options=[\n",
-    "        \"TASK_CONFIG.TASK.ACTIONS.ARM_ACTION.GRIP_CONTROLLER\",\n",
+    "        \"TASK.ACTIONS.ARM_ACTION.GRIP_CONTROLLER\",\n",
     "        \"SuctionGraspAction\",\n",
     "    ],\n",
     ")\n",

examples/tutorials/nb_python/Habitat2_Quickstart.py

@@ -70,14 +70,14 @@
 import numpy as np
 
 import habitat
-import habitat_baselines.utils.gym_definitions as habitat_gym
+import habitat.utils.gym_definitions as habitat_gym
 from habitat.core.embodied_task import Measure
 from habitat.core.registry import registry
 from habitat.core.simulator import Sensor, SensorTypes
 from habitat.tasks.rearrange.rearrange_sensors import RearrangeReward
 from habitat.tasks.rearrange.rearrange_task import RearrangeTask
+from habitat.utils.render_wrapper import overlay_frame
 from habitat.utils.visualizations.utils import observations_to_image
-from habitat_baselines.utils.render_wrapper import overlay_frame
 from habitat_sim.utils import viz_utils as vut
 
 # Quiet the Habitat simulator logging

examples/tutorials/nb_python/habitat2_gym_tutorial.py

@@ -65,7 +65,7 @@
 import gym
 
 # flake8: noqa
-import habitat_baselines.utils.gym_definitions
+import habitat.utils.gym_definitions
 
 # %% [markdown]
 # # Simple Example
@@ -136,7 +136,7 @@
 env = gym.make(
     "HabitatPick-v0",
     override_options=[
-        "TASK_CONFIG.TASK.ACTIONS.ARM_ACTION.GRIP_CONTROLLER",
+        "TASK.ACTIONS.ARM_ACTION.GRIP_CONTROLLER",
         "SuctionGraspAction",
     ],
 )