GP-net: Grasp Proposal for Mobile Manipulators

This dataset includes all trained models and data to use and replicate our work for GP-net (under review for ICRA 2023). The pre-trained model for using GP-net on a PAL TIAGo mobile manipulator with a parallel jaw gripper is available in GP-net_model_pretrained.zip. It can be used with our gpnet-ros repository. If you want to use our dataset and train an alternative network architecture from scratch, use GP-net_training_data.zip. You can use it with the GP-net repository to train models. If you want to run simulation experiments with your trained model, download GP-net_simulation_data.zip for the URDFs of the test objects and pal gripper. If you want to generate a new dataset for an alternative gripper, you can use the gpnet-data repository, which is based on DexNet2.0. The docker image to run the code is available here under gpnet-data_docker_image.tar

本数据集涵盖了用于使用并复现我们提交至ICRA 2023进行评审的GP-net研究工作所需的全部训练模型与数据。针对搭载平行夹爪（parallel jaw gripper）的PAL TIAGo移动操作臂（PAL TIAGo mobile manipulator）的GP-net预训练模型，可在GP-net_model_pretrained.zip中获取，可配合我们的gpnet-ros代码库使用。若您希望使用本数据集并从零开始训练替代网络架构，请使用GP-net_training_data.zip，该数据集可配合GP-net代码库完成模型训练。若您希望使用训练得到的模型运行仿真实验，请下载GP-net_simulation_data.zip以获取测试物体与PAL夹爪的统一机器人描述格式（Unified Robot Description Format，URDF）文件。若您希望为自定义夹爪生成全新的数据集，可使用基于DexNet2.0（DexNet2.0）的gpnet-data代码库。可运行该代码的Docker镜像（Docker image）已在此处以gpnet-data_docker_image.tar的形式提供。

搜集汇总

背景与挑战

背景概述

该数据集包含移动机械手抓取提议的预训练模型、训练数据和仿真数据，支持PAL TIAGo等机械手的抓取任务研究，并提供基于DexNet2.0的数据生成工具。