unimodals.gentle_push package

Submodules

unimodals.gentle_push.head module

Implements gentle_push head module.

Taken from https://github.com/brentyi/multimodalfilter/blob/master/crossmodal/push_models/lstm.py

class unimodals.gentle_push.head.GentlePushLateLSTM(input_size, hidden_size)

Bases: Module

Implements Gentle Push’s Late LSTM model.

__init__(input_size, hidden_size)

Instantiate GentlePushLateLSTM Module.

Parameters:

• input_size (int) – Input dimension

• hidden_size (int) – Hidden dimension

forward(x)

Apply GentlePushLateLSTM to Model Input.

Parameters:

x (torch.Tensor) – Model Input

Returns:

Model Output

Return type:

torch.Tensor

class unimodals.gentle_push.head.Head(units: int = 64)

Bases: Module

Implements Gentle Push’s Head module.

__init__(units: int = 64)

Instantiates Head module.

Parameters:

units (int, optional) – Number of layers in LSTM. Defaults to 64.

forward(fused_features)

Apply Head module to Layer Inputs.

Parameters:

fused_features (torch.Tensor) – Layer Inputs

Returns:

Layer Outputs

Return type:

torch.Tensor

unimodals.gentle_push.layers module

Implements sub-layers for gentle_push model.

Taken from https://github.com/brentyi/multimodalfilter/blob/master/crossmodal/push_models/layers.py.

unimodals.gentle_push.layers.control_layers(units: int) → Module

Create a control command encoder block.

Parameters:

units (int) – # of hidden units in network layers.

Returns:

Encoder block.

Return type:

nn.Module

unimodals.gentle_push.layers.observation_image_layers(units: int, spanning_avg_pool: bool = False) → Module

Create an image encoder block.

Parameters:

units (int) – # of hidden units in network layers.

Returns:

Encoder block.

Return type:

nn.Module

unimodals.gentle_push.layers.observation_pos_layers(units: int) → Module

Create an end effector position encoder block.

Parameters:

units (int) – # of hidden units in network layers.

Returns:

Encoder block.

Return type:

nn.Module

unimodals.gentle_push.layers.observation_sensors_layers(units: int) → Module

Create an F/T sensor encoder block.

Parameters:

units (int) – # of hidden units in network layers.

Returns:

Encoder block.

Return type:

nn.Module

Module contents