Skip to content

env_base

Environment

Bases: gym.Env, GymObservable, Recreatable

Core environment class that handles loading scene, robot(s), and task, following OpenAI Gym interface.

Source code in envs/env_base.py
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
class Environment(gym.Env, GymObservable, Recreatable):
    """
    Core environment class that handles loading scene, robot(s), and task, following OpenAI Gym interface.
    """
    def __init__(
        self,
        configs,
        action_timestep=1 / 60.0,
        physics_timestep=1 / 60.0,
        device=None,
        automatic_reset=False,
    ):
        """
        Args:
            configs (str or dict or list of str or dict): config_file path(s) or raw config dictionaries.
                If multiple configs are specified, they will be merged sequentially in the order specified.
                This allows procedural generation of a "full" config from small sub-configs.
            action_timestep (float): environment executes action per action_timestep second
            physics_timestep: physics timestep for physx
            device (None or str): specifies the device to be used if running on the gpu with torch backend
            automatic_reset (bool): whether to automatic reset after an episode finishes
        """
        # Call super first
        super().__init__()

        # Store settings and other initialized values
        self._automatic_reset = automatic_reset
        self.action_timestep = action_timestep

        # Initialize other placeholders that will be filled in later
        self._initial_pos_z_offset = None                   # how high to offset object placement to account for one action step of dropping
        self._task = None
        self._loaded = None
        self._current_episode = 0

        # Variables reset at the beginning of each episode
        self._current_step = 0

        # Convert config file(s) into a single parsed dict
        configs = configs if isinstance(configs, list) or isinstance(configs, tuple) else [configs]

        # Initial default config
        self.config = self.default_config

        # Merge in specified configs
        for config in configs:
            merge_nested_dicts(base_dict=self.config, extra_dict=parse_config(config), inplace=True)

        # Set the simulator settings
        og.sim.set_simulation_dt(physics_dt=physics_timestep, rendering_dt=action_timestep)
        og.sim.viewer_width = self.render_config["viewer_width"]
        og.sim.viewer_height = self.render_config["viewer_height"]
        og.sim.device = device

        # Load this environment
        self.load()

    def reload(self, configs, overwrite_old=True):
        """
        Reload using another set of config file(s).
        This allows one to change the configuration and hot-reload the environment on the fly.

        Args:
            configs (str or list of str): config_file path(s). If multiple configs are specified, they will
                be merged sequentially in the order specified. This allows procedural generation of a "full" config from
                small sub-configs.
            overwrite_old (bool): If True, will overwrite the internal self.config with @configs. Otherwise, will
                merge in the new config(s) into the pre-existing one. Setting this to False allows for minor
                modifications to be made without having to specify entire configs during each reload.
        """
        # Convert config file(s) into a single parsed dict
        configs = [configs] if isinstance(configs, str) else configs

        # Initial default config
        new_config = self.default_config

        # Merge in specified configs
        for config in configs:
            merge_nested_dicts(base_dict=new_config, extra_dict=parse_config(config), inplace=True)

        # Either merge in or overwrite the old config
        if overwrite_old:
            self.config = new_config
        else:
            merge_nested_dicts(base_dict=self.config, extra_dict=new_config, inplace=True)

        # Load this environment again
        self.load()

    def reload_model(self, scene_model):
        """
        Reload another scene model.
        This allows one to change the scene on the fly.

        Args:
            scene_model (str): new scene model to load (eg.: Rs_int)
        """
        self.scene_config["model"] = scene_model
        self.load()

    def _load_variables(self):
        """
        Load variables from config
        """
        # Store additional variables after config has been loaded fully
        self._initial_pos_z_offset = self.env_config["initial_pos_z_offset"]

        # Reset bookkeeping variables
        self._reset_variables()
        self._current_episode = 0           # Manually set this to 0 since resetting actually increments this

        # - Potentially overwrite the USD entry for the scene if none is specified and we're online sampling -

        # Make sure the requested scene is valid
        scene_type = self.scene_config["type"]
        assert_valid_key(key=scene_type, valid_keys=REGISTERED_SCENES, name="scene type")

        # If we're using a BehaviorTask, we may load a pre-cached scene configuration
        if self.task_config["type"] == "BehaviorTask":
            scene_instance, scene_file = self.scene_config["scene_instance"], self.scene_config["scene_file"]
            if scene_file is None and scene_instance is None and not self.task_config["online_object_sampling"]:
                scene_instance = "{}_task_{}_{}_{}_fixed_furniture_template".format(
                    self.scene_config["scene_model"],
                    self.task_config["activity_name"],
                    self.task_config["activity_definition_id"],
                    self.task_config["activity_instance_id"],
                )
            # Update the value in the scene config
            self.scene_config["scene_instance"] = scene_instance

        # - Additionally run some sanity checks on these values -

        # Check to make sure our z offset is valid -- check that the distance travelled over 1 action timestep is
        # less than the offset we set (dist = 0.5 * gravity * (t^2))
        drop_distance = 0.5 * 9.8 * (self.action_timestep ** 2)
        assert drop_distance < self._initial_pos_z_offset, "initial_pos_z_offset is too small for collision checking"

    def _load_task(self):
        """
        Load task
        """
        # Sanity check task to make sure it's valid
        task_type = self.task_config["type"]
        assert_valid_key(key=task_type, valid_keys=REGISTERED_TASKS, name="task type")

        # Grab the kwargs relevant for the specific task and create the task
        self._task = create_class_from_registry_and_config(
            cls_name=self.task_config["type"],
            cls_registry=REGISTERED_TASKS,
            cfg=self.task_config,
            cls_type_descriptor="task",
        )

        assert og.sim.is_stopped(), "sim should be stopped when load_task starts"
        og.sim.play()

        # Load task. Should load additinal task-relevant objects and configure the scene into its default initial state
        self._task.load(env=self)

        # Update the initial scene state
        self.scene.update_initial_state()

        og.sim.stop()

    def _load_scene(self):
        """
        Load the scene and robot specified in the config file.
        """
        # Create the scene from our scene config
        scene = create_class_from_registry_and_config(
            cls_name=self.scene_config["type"],
            cls_registry=REGISTERED_SCENES,
            cfg=self.scene_config,
            cls_type_descriptor="scene",
        )
        og.sim.import_scene(scene)

    def _load_robots(self):
        """
        Load robots into the scene
        """
        # Only actually load robots if no robot has been imported from the scene loading directly yet
        if len(self.scene.robots) == 0:
            # Iterate over all robots to generate in the robot config
            for i, robot_config in enumerate(self.robots_config):
                # Add a name for the robot if necessary
                if "name" not in robot_config:
                    robot_config["name"] = f"robot{i}"
                # Set prim path
                robot_config["prim_path"] = f"/World/{robot_config['name']}"
                # Make sure robot exists, grab its corresponding kwargs, and create / import the robot
                robot = create_class_from_registry_and_config(
                    cls_name=robot_config["type"],
                    cls_registry=REGISTERED_ROBOTS,
                    cfg=robot_config,
                    cls_type_descriptor="robot",
                )
                # Import the robot into the simulator
                og.sim.import_object(robot)

    def _load_objects(self):
        """
        Load any additional custom objects into the scene
        """
        for i, obj_config in enumerate(self.objects_config):
            # Add a name for the object if necessary
            if "name" not in obj_config:
                obj_config["name"] = f"obj{i}"
            # Set prim path if not specified
            if "prim_path" not in obj_config:
                obj_config["prim_path"] = f"/World/{obj_config['name']}"
            # Pop the desired position and orientation
            position, orientation = obj_config.pop("position", None), obj_config.pop("orientation", None)
            # Make sure robot exists, grab its corresponding kwargs, and create / import the robot
            obj = create_class_from_registry_and_config(
                cls_name=obj_config["type"],
                cls_registry=REGISTERED_OBJECTS,
                cfg=obj_config,
                cls_type_descriptor="object",
            )
            # Import the robot into the simulator and set the pose
            og.sim.import_object(obj)
            obj.set_position_orientation(position=position, orientation=orientation)

    def _load_observation_space(self):
        # Grab robot(s) and task obs spaces
        obs_space = OrderedDict()

        for robot in self.robots:
            # Load the observation space for the robot
            obs_space[robot.name] = robot.load_observation_space()

        # Also load the task obs space
        obs_space["task"] = self._task.load_observation_space()
        return obs_space

    def _load_action_space(self):
        """
        Load action space for each robot
        """
        self.action_space = gym.spaces.Dict({robot.name: robot.action_space for robot in self.robots})

    def load(self):
        """
        Load the scene and robot specified in the config file.
        """
        # This environment is not loaded
        self._loaded = False

        # Load config variables
        self._load_variables()

        # Load the scene, robots, and task
        self._load_scene()
        self._load_robots()
        self._load_objects()
        self._load_task()

        og.sim.play()
        self.reset()

        # Load the obs / action spaces
        self.load_observation_space()
        self._load_action_space()

        # Denote that the scene is loaded
        self._loaded = True

    def close(self):
        """
        Clean up the environment and shut down the simulation.
        """
        og.sim.close()

    def get_obs(self):
        """
        Get the current environment observation.

        Returns:
            OrderedDict: Keyword-mapped observations, which are possibly nested
        """
        obs = OrderedDict()

        # Grab all observations from each robot
        for robot in self.robots:
            obs[robot.name] = robot.get_obs()

        # Add task observations
        obs["task"] = self._task.get_obs(env=self)

        return obs

    def _populate_info(self, info):
        """
        Populate info dictionary with any useful information.

        Args:
            info (dict): Information dictionary to populate

        Returns:
            dict: Information dictionary with added info
        """
        info["episode_length"] = self._current_step

    def step(self, action):
        """
        Apply robot's action and return the next state, reward, done and info,
        following OpenAI Gym's convention

        Args:
            action (gym.spaces.Dict or dict or np.array): robot actions. If a dict is specified, each entry should
                map robot name to corresponding action. If a np.array, it should be the flattened, concatenated set
                of actions

        Returns:
            4-tuple:
                - OrderedDict: state, i.e. next observation
                - float: reward, i.e. reward at this current timestep
                - bool: done, i.e. whether this episode is terminated
                - OrderedDict: info, i.e. dictionary with any useful information
        """
        # If the action is not a dictionary, convert into a dictionary
        if not isinstance(action, dict) and not isinstance(action, gym.spaces.Dict):
            action_dict = OrderedDict()
            idx = 0
            for robot in self.robots:
                action_dim = robot.action_dim
                action_dict[robot.name] = action[idx: idx + action_dim]
                idx += action_dim
        else:
            # Our inputted action is the action dictionary
            action_dict = action

        # Iterate over all robots and apply actions
        for robot in self.robots:
            robot.apply_action(action_dict[robot.name])

        # Run simulation step
        og.sim.step()

        # Grab observations
        obs = self.get_obs()

        # Grab reward, done, and info, and populate with internal info
        reward, done, info = self.task.step(self, action)
        self._populate_info(info)

        if done and self._automatic_reset:
            # Add lost observation to our information dict, and reset
            info["last_observation"] = obs
            obs = self.reset()

        # Increment step
        self._current_step += 1

        return obs, reward, done, info

    def _reset_variables(self):
        """
        Reset bookkeeping variables for the next new episode.
        """
        self._current_episode += 1
        self._current_step = 0

    # TODO: Match super class signature?
    def reset(self):
        """
        Reset episode.
        """
        # Stop and restart the simulation
        og.sim.stop()
        og.sim.play()

        # Reset the task
        self.task.reset(self)

        # Reset internal variables
        self._reset_variables()

        # Run a single simulator step to make sure we can grab updated observations
        og.sim.step()

        # Grab and return observations
        obs = self.get_obs()

        if self.observation_space is not None and not self.observation_space.contains(obs):
            # Print out all observations for all robots and task
            for robot in self.robots:
                for key, value in self.observation_space[robot.name].items():
                    logging.error(("obs_space", key, value.dtype, value.shape))
                    logging.error(("obs", key, obs[robot.name][key].dtype, obs[robot.name][key].shape))
            for key, value in self.observation_space["task"].items():
                logging.error(("obs_space", key, value.dtype, value.shape))
                logging.error(("obs", key, obs["task"][key].dtype, obs["task"][key].shape))
            raise ValueError("Observation space does not match returned observations!")

        return obs

    @property
    def episode_steps(self):
        """
        Returns:
            int: Current number of steps in episode
        """
        return self._current_step

    @property
    def initial_pos_z_offset(self):
        """
        Returns:
            float: how high to offset object placement to test valid pose & account for one action step of dropping
        """
        return self._initial_pos_z_offset

    @property
    def task(self):
        """
        Returns:
            BaseTask: Active task instance
        """
        return self._task

    @property
    def scene(self):
        """
        Returns:
            Scene: Active scene in this environment
        """
        return og.sim.scene

    @property
    def robots(self):
        """
        Returns:
            list of BaseRobot: Robots in the current scene
        """
        return self.scene.robots

    @property
    def env_config(self):
        """
        Returns:
            dict: Environment-specific configuration kwargs
        """
        return self.config["env"]

    @property
    def render_config(self):
        """
        Returns:
            dict: Render-specific configuration kwargs
        """
        return self.config["render"]

    @property
    def scene_config(self):
        """
        Returns:
            dict: Scene-specific configuration kwargs
        """
        return self.config["scene"]

    @property
    def robots_config(self):
        """
        Returns:
            dict: Robot-specific configuration kwargs
        """
        return self.config["robots"]

    @property
    def objects_config(self):
        """
        Returns:
            dict: Object-specific configuration kwargs
        """
        return self.config["objects"]

    @property
    def task_config(self):
        """
        Returns:
            dict: Task-specific configuration kwargs
        """
        return self.config["task"]

    @property
    def default_config(self):
        """
        Returns:
            dict: Default configuration for this environment. May not be fully specified (i.e.: still requires @config
                to be specified during environment creation)
        """
        return {
            # Environment kwargs
            "env": {
                "initial_pos_z_offset": 0.1,
            },

            # Rendering kwargs
            "render": {
                "viewer_width": 1280,
                "viewer_height": 720,
            },

            # Scene kwargs
            "scene": {
                # Traversibility map kwargs
                "waypoint_resolution": 0.2,
                "num_waypoints": 10,
                "build_graph": False,
                "trav_map_resolution": 0.1,
                "trav_map_erosion": 2,
                "trav_map_with_objects": True,
                "scene_instance": None,
                "scene_file": None,
            },

            # Robot kwargs
            "robots": [],   # no robots by default

            # Object kwargs
            "objects": [],  # no objects by default

            # Task kwargs
            "task": {
                "type": "DummyTask",

                # If we're using a BehaviorTask
                "activity_definition_id": 0,
                "activity_instance_id": 0,
            }
        }

default_config property

Returns:

Name Type Description
dict

Default configuration for this environment. May not be fully specified (i.e.: still requires @config to be specified during environment creation)

env_config property

Returns:

Name Type Description
dict

Environment-specific configuration kwargs

episode_steps property

Returns:

Name Type Description
int

Current number of steps in episode

initial_pos_z_offset property

Returns:

Name Type Description
float

how high to offset object placement to test valid pose & account for one action step of dropping

objects_config property

Returns:

Name Type Description
dict

Object-specific configuration kwargs

render_config property

Returns:

Name Type Description
dict

Render-specific configuration kwargs

robots property

Returns:

Type Description

list of BaseRobot: Robots in the current scene

robots_config property

Returns:

Name Type Description
dict

Robot-specific configuration kwargs

scene property

Returns:

Name Type Description
Scene

Active scene in this environment

scene_config property

Returns:

Name Type Description
dict

Scene-specific configuration kwargs

task property

Returns:

Name Type Description
BaseTask

Active task instance

task_config property

Returns:

Name Type Description
dict

Task-specific configuration kwargs

__init__(configs, action_timestep=1 / 60.0, physics_timestep=1 / 60.0, device=None, automatic_reset=False)

Parameters:

Name Type Description Default
configs str or dict or list of str or dict

config_file path(s) or raw config dictionaries. If multiple configs are specified, they will be merged sequentially in the order specified. This allows procedural generation of a "full" config from small sub-configs.

required
action_timestep float

environment executes action per action_timestep second

1 / 60.0
physics_timestep

physics timestep for physx

1 / 60.0
device None or str

specifies the device to be used if running on the gpu with torch backend

None
automatic_reset bool

whether to automatic reset after an episode finishes

False
Source code in envs/env_base.py
def __init__(
    self,
    configs,
    action_timestep=1 / 60.0,
    physics_timestep=1 / 60.0,
    device=None,
    automatic_reset=False,
):
    """
    Args:
        configs (str or dict or list of str or dict): config_file path(s) or raw config dictionaries.
            If multiple configs are specified, they will be merged sequentially in the order specified.
            This allows procedural generation of a "full" config from small sub-configs.
        action_timestep (float): environment executes action per action_timestep second
        physics_timestep: physics timestep for physx
        device (None or str): specifies the device to be used if running on the gpu with torch backend
        automatic_reset (bool): whether to automatic reset after an episode finishes
    """
    # Call super first
    super().__init__()

    # Store settings and other initialized values
    self._automatic_reset = automatic_reset
    self.action_timestep = action_timestep

    # Initialize other placeholders that will be filled in later
    self._initial_pos_z_offset = None                   # how high to offset object placement to account for one action step of dropping
    self._task = None
    self._loaded = None
    self._current_episode = 0

    # Variables reset at the beginning of each episode
    self._current_step = 0

    # Convert config file(s) into a single parsed dict
    configs = configs if isinstance(configs, list) or isinstance(configs, tuple) else [configs]

    # Initial default config
    self.config = self.default_config

    # Merge in specified configs
    for config in configs:
        merge_nested_dicts(base_dict=self.config, extra_dict=parse_config(config), inplace=True)

    # Set the simulator settings
    og.sim.set_simulation_dt(physics_dt=physics_timestep, rendering_dt=action_timestep)
    og.sim.viewer_width = self.render_config["viewer_width"]
    og.sim.viewer_height = self.render_config["viewer_height"]
    og.sim.device = device

    # Load this environment
    self.load()

close()

Clean up the environment and shut down the simulation.

Source code in envs/env_base.py
def close(self):
    """
    Clean up the environment and shut down the simulation.
    """
    og.sim.close()

get_obs()

Get the current environment observation.

Returns:

Name Type Description
OrderedDict

Keyword-mapped observations, which are possibly nested

Source code in envs/env_base.py
def get_obs(self):
    """
    Get the current environment observation.

    Returns:
        OrderedDict: Keyword-mapped observations, which are possibly nested
    """
    obs = OrderedDict()

    # Grab all observations from each robot
    for robot in self.robots:
        obs[robot.name] = robot.get_obs()

    # Add task observations
    obs["task"] = self._task.get_obs(env=self)

    return obs

load()

Load the scene and robot specified in the config file.

Source code in envs/env_base.py
def load(self):
    """
    Load the scene and robot specified in the config file.
    """
    # This environment is not loaded
    self._loaded = False

    # Load config variables
    self._load_variables()

    # Load the scene, robots, and task
    self._load_scene()
    self._load_robots()
    self._load_objects()
    self._load_task()

    og.sim.play()
    self.reset()

    # Load the obs / action spaces
    self.load_observation_space()
    self._load_action_space()

    # Denote that the scene is loaded
    self._loaded = True

reload(configs, overwrite_old=True)

Reload using another set of config file(s). This allows one to change the configuration and hot-reload the environment on the fly.

Parameters:

Name Type Description Default
configs str or list of str

config_file path(s). If multiple configs are specified, they will be merged sequentially in the order specified. This allows procedural generation of a "full" config from small sub-configs.

required
overwrite_old bool

If True, will overwrite the internal self.config with @configs. Otherwise, will merge in the new config(s) into the pre-existing one. Setting this to False allows for minor modifications to be made without having to specify entire configs during each reload.

True
Source code in envs/env_base.py
def reload(self, configs, overwrite_old=True):
    """
    Reload using another set of config file(s).
    This allows one to change the configuration and hot-reload the environment on the fly.

    Args:
        configs (str or list of str): config_file path(s). If multiple configs are specified, they will
            be merged sequentially in the order specified. This allows procedural generation of a "full" config from
            small sub-configs.
        overwrite_old (bool): If True, will overwrite the internal self.config with @configs. Otherwise, will
            merge in the new config(s) into the pre-existing one. Setting this to False allows for minor
            modifications to be made without having to specify entire configs during each reload.
    """
    # Convert config file(s) into a single parsed dict
    configs = [configs] if isinstance(configs, str) else configs

    # Initial default config
    new_config = self.default_config

    # Merge in specified configs
    for config in configs:
        merge_nested_dicts(base_dict=new_config, extra_dict=parse_config(config), inplace=True)

    # Either merge in or overwrite the old config
    if overwrite_old:
        self.config = new_config
    else:
        merge_nested_dicts(base_dict=self.config, extra_dict=new_config, inplace=True)

    # Load this environment again
    self.load()

reload_model(scene_model)

Reload another scene model. This allows one to change the scene on the fly.

Parameters:

Name Type Description Default
scene_model str

new scene model to load (eg.: Rs_int)

required
Source code in envs/env_base.py
def reload_model(self, scene_model):
    """
    Reload another scene model.
    This allows one to change the scene on the fly.

    Args:
        scene_model (str): new scene model to load (eg.: Rs_int)
    """
    self.scene_config["model"] = scene_model
    self.load()

reset()

Reset episode.

Source code in envs/env_base.py
def reset(self):
    """
    Reset episode.
    """
    # Stop and restart the simulation
    og.sim.stop()
    og.sim.play()

    # Reset the task
    self.task.reset(self)

    # Reset internal variables
    self._reset_variables()

    # Run a single simulator step to make sure we can grab updated observations
    og.sim.step()

    # Grab and return observations
    obs = self.get_obs()

    if self.observation_space is not None and not self.observation_space.contains(obs):
        # Print out all observations for all robots and task
        for robot in self.robots:
            for key, value in self.observation_space[robot.name].items():
                logging.error(("obs_space", key, value.dtype, value.shape))
                logging.error(("obs", key, obs[robot.name][key].dtype, obs[robot.name][key].shape))
        for key, value in self.observation_space["task"].items():
            logging.error(("obs_space", key, value.dtype, value.shape))
            logging.error(("obs", key, obs["task"][key].dtype, obs["task"][key].shape))
        raise ValueError("Observation space does not match returned observations!")

    return obs

step(action)

Apply robot's action and return the next state, reward, done and info, following OpenAI Gym's convention

Parameters:

Name Type Description Default
action gym.spaces.Dict or dict or np.array

robot actions. If a dict is specified, each entry should map robot name to corresponding action. If a np.array, it should be the flattened, concatenated set of actions

required

Returns:

Type Description

4-tuple: - OrderedDict: state, i.e. next observation - float: reward, i.e. reward at this current timestep - bool: done, i.e. whether this episode is terminated - OrderedDict: info, i.e. dictionary with any useful information

Source code in envs/env_base.py
def step(self, action):
    """
    Apply robot's action and return the next state, reward, done and info,
    following OpenAI Gym's convention

    Args:
        action (gym.spaces.Dict or dict or np.array): robot actions. If a dict is specified, each entry should
            map robot name to corresponding action. If a np.array, it should be the flattened, concatenated set
            of actions

    Returns:
        4-tuple:
            - OrderedDict: state, i.e. next observation
            - float: reward, i.e. reward at this current timestep
            - bool: done, i.e. whether this episode is terminated
            - OrderedDict: info, i.e. dictionary with any useful information
    """
    # If the action is not a dictionary, convert into a dictionary
    if not isinstance(action, dict) and not isinstance(action, gym.spaces.Dict):
        action_dict = OrderedDict()
        idx = 0
        for robot in self.robots:
            action_dim = robot.action_dim
            action_dict[robot.name] = action[idx: idx + action_dim]
            idx += action_dim
    else:
        # Our inputted action is the action dictionary
        action_dict = action

    # Iterate over all robots and apply actions
    for robot in self.robots:
        robot.apply_action(action_dict[robot.name])

    # Run simulation step
    og.sim.step()

    # Grab observations
    obs = self.get_obs()

    # Grab reward, done, and info, and populate with internal info
    reward, done, info = self.task.step(self, action)
    self._populate_info(info)

    if done and self._automatic_reset:
        # Add lost observation to our information dict, and reset
        info["last_observation"] = obs
        obs = self.reset()

    # Increment step
    self._current_step += 1

    return obs, reward, done, info