Touch Detection

The TouchDetect task predicts if there is an object in contact with the surface of the sensor.

Available Pre-Trained Models

The following pre-trained models are available for use with the TouchDetect task:

Dataset	Model Name	Accuracy
DIGIT	touchdetect_	xx.yy
GelSight	touchdetect_	xx.yy
OmniTact	touchdetect_	xx.yy
Joint (DIGIT, Gelsight, OmniTact)	joint_touchdetect_	xx.yy

Usage

Initialize the TouchDetect task with a sensor and pre-trained model,

touch_detect = TouchDetect(DigitSensor, zoo_model="touchdetect_resnet18")

Calling touch_detect on an input sample predicts with certainty if there is an object touching the sensor surface.

is_touching, certainty = touch_detect(sample)

Normalization

The TouchDetect task loads default transform and normalization values from pytouch.models.touch_detect.TouchDetectModelDefaults and is suitable for any pre-trained model from the TorchVision package.

For custom models provide a custom class when initializing TouchDetect in the format of:

touch_detect = TouchDetect(model=my_custom_model, defaults=MyTouchDetectValues)

Where MyTouchDetectValues are in the following format,

class MyTouchDetectValues:

SCALES = [64, 64]

MEANS = [0.123, 0.123, 0.123]

STDS = [0.123, 0.123, 0.123]

CLASSES = 2