Train a neural network by using video data from GelSight touch sensor to detect slip when a robot lift an object

Among many robotic tasks that require assistance of tactile sensing, the most important task is to detect whether the robot has safely grasped an object. Slip, a common grasp failure, will occur when the gripping force is not large enough or grasping position is inappropriate. GleSight sensor provides the geometry deformation, shear and toque forces of the contact surface when robot grasps something. By using robot arm equipped with GelSight sensor to lift a large amount of daily objects, we collect many video data for different trials. A Pre-trained VGG-net followed by a LSTM is used as the neural network architecture and the output of the neural network is whether slip occurs or not.

Research Areas