Get the velocity of a named body in the simulator world reference frame

Build a new physics simulation environment

Pick a new goal button, maybe with resampling due to hazards

Rejection sample a placement of objects to find a layout.

Build a dict of placements. Happens once during __init__.

Create a world_config from our own config

Tick the buttons resampling timer

Check that a benchmark env has the right lidar obs for the objects in scene

Run a single environment for a single episode

Check that all the names in the mujoco model are the same for different envs

Check that a single environment is seed-stable at init

Collects rollouts and computes advantages. Runs several environments concurrently. The next actions are computed in a batch mode for

Return the egocentric XY vector to a position from the robot

Return true if the current goal is met this step

Create a wrapped, monitored SubprocVecEnv for Atari.

Natural lidar casts rays based on the ego-frame of the robot. Rays are circularly projected from the robot body origin around

Return a robot-centric lidar observation of a list of positions. Lidar is a set of bins around the robot (divided evenly in a circle

Parse a config dict - see self.DEFAULT for description

This function receives the LTL formulas and convert them into inputs for a GNN

This function receives the LTL formulas and convert them into inputs for an RNN

Convert Quaternion to a 3x3 Rotation Matrix using mujoco

From quaternion, extract z_{ground} dot z_{body}

Build a new sim from a model if the model changed

Render the environment to the screen

Render the environment to the screen

Render a radial area in the environment

Render a radial area in the environment

This function resets the world and collects the first observation.

Calculate the dense component of reward. Call exactly once per step

Get the velocity of the robot in the simulator world reference frame

Sample a new goal position and return True, else False if sample rejected

Sample a single layout, returning True if successful, else False.

Set mocap object positions before a physics step is executed

This function executes an action in the environment

Convert an angle (in radians) to a unit vector in that angle around Z

Update layout dictionary with new places of objects

Removes the LTL formula from an LTLEnv It is useful to check the performance of off-the-shelf agents