rofunc.learning.RofuncRL.tasks.isaacgymenv.furniture

Contents

`rofunc.learning.RofuncRL.tasks.isaacgymenv.furniture`#

Define base environment class FurnitureEnv.

1. Module Contents#

1.1. Classes#

FurnitureTask

Base class for IKEA furniture assembly environment.

1.2. Data#

NEW_CONTROLLERS

1.3. API#

rofunc.learning.RofuncRL.tasks.isaacgymenv.furniture.NEW_CONTROLLERS = ['position', 'position_orientation', 'joint_impedance', 'joint_torque', 'joint_velocity']#

class rofunc.learning.RofuncRL.tasks.isaacgymenv.furniture.FurnitureTask(config)#

Base class for IKEA furniture assembly environment.

Initialization

Initializes class with the configuration.

name = 'furniture'#

update_config(config)#: Updates private member variables with @config dictionary.

set_subtask(subtask, num_connects=None)#: Simply sets @self._preassembled to [0, 1, …, @subtask].

num_subtask()#

property observation_space#: Returns dict where keys are ob names and values are dimensions.

property state_space#: Returns mujoco state space.

abstract property dof#: Returns the DoF of the robot (with out grippers).

property max_episode_steps#: Returns maximum number of steps in an episode.

property action_size#: Returns size of the action space.

property action_space#: Returns action space.

set_max_episode_steps(max_episode_steps)#

set_init_qpos(init_qpos)#

reset(furniture_id=None, background=None)#: Takes in a furniture_id, and background string. Resets the environment, viewer, and internal variables. Returns the initial observation.

step(action)#: Computes the next environment state given @action. Stores env state for demonstration if needed. Returns observation dict, reward float, done bool, and info dict.

render(mode='human')#: Renders the environment. If mode is rgb_array, we render the pixels. The pixels can be rgb, depth map, segmentation map If the mode is human, we render to the MuJoCo viewer, or for unity, do nothing since rendering is handled by Unity.

close()#: Cleans up the environment

get_env_state()#: Returns current qpos and qvel.

set_env_state(given_state)#

key_callback(window, key, scancode, action, mods)#: Key listener for MuJoCo viewer

key_input_unity()#: Key input for unity If adding new keys, make sure to add keys to whitelist in MJTCPInterace.cs

resize_key_input_unity()#: Key input for unity If adding new keys, make sure to add keys to whitelist in MJTCPInterace.cs

run_demo(config=None)#: Since we save all states, just play back states

get_vr_input(controller)#

run_vr(config=None)#: Runs the environment with HTC Vive support

run_manual(config=None)#: Run the environment under manual (keyboard) control

run_demo_actions(config=None)#: Play the stored actions in demonstration

run_resizer(config=None)#: Run a resizing program in unity for adjusting furniture size in xml

run_img(config=None)#: Run a resizing program in unity for adjusting furniture size in xml

initialize_time()#: Initializes the time constants used for simulation.

set_state(qpos, qvel)#: Sets the qpos and qvel of the MuJoCo sim

on_collision(ref_name, body_name=None)#: Checks if there is collision

pose_in_base_from_name(name)#: A helper function that takes in a named data field and returns the pose of that object in the base frame.