Created: August 22, 2022

Tags: Hardness, Shore scale, LSTM, Gelsight

Link: https://arxiv.org/abs/1704.03955

Status: Reading

What?

Estimating hardness of an object is important for robot grasping. Current sensors do not provide such information.

Why?

After touching the object, if the robot can feel how hard the object is, it can manipulate the object better by applying (predefined) force and torque.

This hardness estimation can also be used for haptic feedback in teleoperated robot grippers. By idenfitying object hardness, specific force-position control can be employed.

How?

Assumption: during contact, soft objects deform more and produce a smoother surface. Hard objects deform less and ridges remain sharp.

It is important to check if this actually happens in our gel, as well !

Under different hardness, the following changes:

• Color intensity
• Marker field ( which indicates contact force through its motion)

Architecture is simple and similar to action recognition pipeline:

1. Represent each frame with CNN features ( CNN embedding) (VGG or ResNet backbone)
2. Connect this to LSTM to model temporal information