Robust and Perceptive Manipulation Through Soft-Bubble Grippers and Monocular Depth

A method comprising:

• pressing a training object against a deformable membrane of a deformable sensor, the deformable sensor comprising an enclosure comprising the deformable membrane, the enclosure configured to be filled with a medium;
• with an imaging sensor disposed within the enclosure, capturing an image of the deformable membrane while the training object contacts the deformable membrane;
• determining a depth value associated with each pixel of the image captured by the imaging sensor;
• and
• storing the image of the deformable membrane and the depth value associated with each pixel of the image as a training example for a neural network;
• and
• using the training example to train the neural network to estimate the depth value associated with each pixel based on the image of the deformable membrane.

A method comprising:

• pressing an object against a deformable membrane of a deformable sensor, the deformable sensor comprising an enclosure comprising the deformable membrane, the enclosure configured to be filled with a medium;
• with an imaging sensor disposed within the enclosure, capturing an image of the deformable membrane while the object contacts the deformable membrane;
• and
• using a trained neural network to determine a depth value associated with each pixel of the image captured by the imaging sensor.