World

A world encompasses the reality the robot experiment takes place, this includes simulations like PyBullet, as well as The Simulation We Inhabit known as reality. Worlds contain RobotInterfaces, ObjectInterfaces, Arenas and SensorInterfaces. They do not include experimental concepts such as tasks and reward functions.

World

class perls2.worlds.world.World(cfg_path)

Abstract base class for Worlds.

A World encompasses all necessary components for managing an experiment in simulation or reality. Worlds generate the appropriate robot, sensor and object interfaces based on the configuration file.

config

Config files contain parameters to create an arena, robot interface, sensor interface and object interface. They also contain specs for learning, simulation and experiment setup.

Type

dict

arena

Manages the sim by loading models for real and sim and for simulations, randomizing objects and sensors parameters

Type

Arena

robot_interface

Communicates with robots and executes robot commands.

Type

RobotInterface

sensor_interface

Retrieves sensor info and executes changes to params

Type

SensorInterface

is_sim

Flag that is true if world is simulation.

Type

bool

abstract reset()

Reset the world to initial state.

Returns

The observation.

abstract step(start=None)

Step simulation forward.

BulletWorld

class perls2.worlds.bullet_world.BulletWorld(config, use_visualizer=False, name=None)

Class for PyBullet worlds.

A BulletWorld will automatically create instances of BulletRobot, BulletCamera and BulletObjectInterfaces based on the config file.

arena

Manages the sim by loading models, and for simulations, randomizing objects and sensors params.

Type

BulletArena

robot_interface

Communicates with robots and executes robot commands.

Type

BulletRobotInterface

camera_interface

Retrieves camera images and executes changes to params (e.g. intrinsics/extrinsics)

Type

BulletCameraInterface

object_interfaces

A dictionary of perls2.BulletObjectInterfaces currently in the simulation. Keys are string unique identifying names for the object.

Type

dict

physics_id

Unique id identifying physics client for Pybullet. Used to connect other interfaces when working with multiple simulations.

Type

int

time_step

float Float defining timestep to increment Pybullet simulation during pybullet.stepSimulation() calls. Set in config file

MAX_STEPS

int Constant for maximum number of steps before terminating epsiode. Set in config file

name

str A name to describe the world (e.g. training or testing)

add_object(path, name, pose, scale=1.0, is_static=False)

Add object to world explicitly.

Parameters

  • path (str) – filepath name to object urdf

  • name (str) – name of object used for dictionary key

  • pose (list) – ((3,),(4,)) pose of the object as [[x, y, z], [qx, qy, qz, w]] orientation as quaternion.

  • scale (double) – scale of object.

  • is_static (bool) – whether object should remain fixed or not.

Returns

object interface added to world.

Return type

object_interface (ObjectInterface)

Examples

objI = self.add_object(‘objects/ycb/013_apple/google_16k/textured.urdf’, ‘013_apple’, [0, 0, 0, 0, 0, 0, 1], 1.0,)

reboot()

Reboot pybullet simulation by clearing all objects, urdfs and resetting sim state.

Warning: this is a slow process and should only be used for restoring state and deterministic applications.

reconnect()

Disconnects and reconnects to new physics engine

remove_object(name)

Remove object from world. :param name: name of the object as stored in objects dictionary. :type name: str

Notes: Removes object from arena as well as objects dictionary.

reset()

Reset the world.

Returns

None

TODO: Should this return something? an error? This needs to be defined by derived class. But all derived reset functions share these steps.

set_pb_physics()

Set physics parameters for pybullet simulation.

set_state(filepath)

Set simulation to .bullet path found in filepath

Parameters

filepath (str) – filepath of .bullet file for saved state.

Returns

None

Examples

env.world.set_state(‘/path/to/state.bullet’)

Notes

To use this function correctly, the same objects / robots must be loaded in the same order. Therefore it is only recommended to use with the same world / config, rather than trying to load an empty world.

step(start=None)

Step the world(simulation) forward.

Parameters

start (float) – timestamp for when env.step was called by policy. Ignored for this class.

Returns

None

Step simulation forward a number of times per action to ensure smoothness when for collisions

visualize(observation, action)

Visualize the action.

Add visual markers to the world (in case of sim) or execute some movements (in case of real) to indicate the action about to be performed.

Parameters

  • observation – The observation of the current step.

  • action – The selected action.

visualize_safenet_boundaries()

Add visualization to pb sim for safenet boundary.

wait_until_stable(linear_velocity_threshold=0.005, angular_velocity_threshold=0.005, check_after_steps=100, min_stable_steps=100, max_steps=30000)

Wait until the objects are stable.

Blocking code that checks if object linear and angular velocity are within respectives thresholds of 0.

Parameters

  • linear_velocity_threshold (float) –

  • angular_velocity_threshold (float) –

  • check_after_steps (int) – min number of steps to wait before checking object state

  • min_stable_step (int) – min number of steps for object to be stable before exiting

  • max_steps (int) – max number of steps to wait for stable object

Returns

None

TODO

  • add check for angular velocity threshold

RealWorld

class perls2.worlds.real_world.RealWorld(config=None, use_visualizer=False, name='DefaultEnv')

Typically, an env will have one robot with one camera as well as logic to have the robot execute a particular task.

reset()

Reset the environment.

Returns

The observation.

step(start=None)

Take a step.

Parameters

start (float) – time.time() timestamp taken from before policy computes action. This is to enforce policy frequency.

Returns: None

Takes a step forward, since this happens naturally in reality, we don’t do anything.