Moves the robot arm so that the gripper center ends up at the requested XYZ-position, with a vertical gripper position. Args

Moves the specified joint group to a joint target. Args: group: String specifying the group to move. target:

Lets the agent execute the action. Args: action: The action to be performed. Returns: observation:

Holds the current position by actuating the joints towards their current target position. Args: duration: Time in ms to

Takes an observation dictionary, transforms it into a normalized tensor of shape (1,4,height,width). The returned tensor will already

Closes the gripper while keeping the arm in a steady position.

3x3 convolution with padding

Uses the controllers get_image_data method to return an top-down image (as a np-array). Args: show: If True, displays th

Adds a circular red marker at the coordinates, dislaying the coordinates as a label. Args: coordinates: List of XYZ-coor

Checks weather each predictions lies within a specified threshold of the ground truth

Initializes all camera parameters that only need to be calculated once.

Allows the user to create custom objects for controlling groups of joints. The method show_model_info can be used to get lists of joi

Returns an action according to the epsilon-greedy policy. Args: state: An observation / state that will be forwarded thr

Returns the RGB and depth images of the provided camera. Args: show: If True displays the images for five seconds or unt

Attempts a grasp at the current location and prints some feedback on weather it was successful

Method for solving simple inverse kinematic problems. This was developed for top down graspig, therefore the solution will be one whe

Example implementaion of a training method, using standard DQN-learning. Samples batches from the replay buffer, feeds them through t

Opens the gripper while keeping the arm in a steady position.

Converts pixel coordinates into world coordinates. Args: pixel_x: X-coordinate in pixel space. pixel_y: Y-co

Method for keeping track of tensorboard metrics. Args: reward: Reward to be added to the list of last 1000 rewards.

Saves the recorded joint values as a .png-file. The values for each joint of the group are put in a seperate subplot. Args:

Creates some basic lists and fill them with initial values. This function is called in the class costructor. The following lists/dict

Converts the depth array delivered by MuJoCo (values between 0 and 1) into actual m values. Args: depth: The depth array

Creates a two dimensional list of joint angles for plotting. Args: group: The group involved in the movement.

Reads and returns the mean and standard deviation values created by 'normalize.py'.

Displays relevant model info for the user, namely bodies, joints, actuators, as well as their IDs and ranges. Also gives info on whic

Test method for trying out tossing of grasped objects.

Args: height: Observation height (in pixels). width: Observation width (in pixels). mem_size: Number of t

Hook for displaying the indices of non-zero gradients. Useful for making sure only the loss of pixels corresponding to selected actio

Prints the function signature and return value

Debug method, simply displays some relevant data at the time of the call.

Always returns the greedy action. For demonstrating learned behaviour. Args: state: An observation / state that will be

TODO: Implement orientation.

Method to perform additional reset steps and return an observation. Gets called in the parent classes reset method.

Legacy method, not used in the current setup. May be used to directly set the joint values to a desired position.

Use this setter for changing .

Prints the runtime of the decorated function.

Takes a XYZ world position and transforms it into pixel coordinates. Mainly implemented for testing the correctness of the camera mat