Data Generation for Learning to Grasp in a Bin-picking Scenario

TL;DR

This paper presents a comprehensive simulation-based data generation method for robotic bin-picking, creating a large dataset with diverse conditions to improve deep learning grasping models.

Contribution

It introduces a scalable simulation framework for generating diverse bin-picking data with ground truth annotations, facilitating data-driven robotic grasping research.

Findings

Generated 100K diverse data samples for bin-picking

Included various environmental conditions like lighting and noise

Made dataset and source code publicly available

Abstract

The rise of deep learning has greatly transformed the pipeline of robotic grasping from model-based approach to data-driven stream. Along this line, a large scale of grasping data either collected from simulation or from real world examples become extremely important. In this paper, we present our recent work on data generation in simulation for a bin-picking scene. 77 objects from the YCB object data sets are used to generate the dataset with PyBullet, where different environment conditions are taken into account including lighting, camera pose, sensor noise and so on. In all, 100K data samples are collected in terms of ground truth segmentation, RGB, 6D pose and point cloud. All the data examples including the source code are made available online.