Slip Detection

The SlipDetect task predicts if an object has slipped while being grasped.

Available Pre-Trained Models

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

Dataset	Model Name	Accuracy
DIGIT	slipdetect_resnet18	xx.yy

Input

Input into the SlipDetect task expects a 12 consecutive frames to predict if slip has occured.

Usage

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

slip_detect = SlipDetect(DigitSensor, zoo_model="slipdetect_resnet18")

Normalization

The SlipDetect task loads default transform and normalization values from pytouch.models.slip_detect.SlipDetectModelDefaults and is suitable for any pre-trained model from the TorchVision package.

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

slip_detect = SlipDetect(model=my_custom_model, defaults=MySlipDetectValues)

Where MySlipDetectValues are in the following format,

class MySlipDetectValues:

SCALES = [64, 64]

MEANS = [0.123, 0.123, 0.123]

STDS = [0.123, 0.123, 0.123]