Self Supervised Deep Learning for Robot Grasping

TL;DR

This paper introduces a self-supervised learning approach for robot grasping, enabling robots to autonomously generate training data, reducing labeling effort, and improving scalability in grasping tasks.

Contribution

It presents a novel self-supervised training setup for CNN-based robot grasping, eliminating the need for manual data labeling and enabling autonomous data collection.

Findings

Robot can autonomously label and collect grasp data.

CNN trained on large autonomous dataset improves grasping performance.

Reduced human bias and effort in data annotation.

Abstract

Learning Based Robot Grasping currently involves the use of labeled data. This approach has two major disadvantages. Firstly, labeling data for grasp points and angles is a strenuous process, so the dataset remains limited. Secondly, human labeling is prone to bias due to semantics. In order to solve these problems we propose a simpler self-supervised robotic setup, that will train a Convolutional Neural Network (CNN). The robot will label and collect the data during the training process. The idea is to make a robot that is less costly, small and easily maintainable in a lab setup. The robot will be trained on a large data set for several hundred hours and then the trained Neural Network can be mapped onto a larger grasping robot.