RLBench

RLBench是由帝国理工学院戴森机器人实验室创建的一个大型机器人学习基准和学习环境，包含100个独特的手工设计任务，涵盖从简单的目标达到和门打开到复杂的多阶段任务。数据集提供多种感知观察，包括rgb、深度和分割掩码，以及机器人本体感受数据。每个任务都配备了无限数量的演示，通过使用基于任务创建时给出的航点的运动规划器实现。RLBench旨在加速视觉引导操作研究领域的进展，特别是少样本学习。数据集的设计考虑了可扩展性，允许用户轻松创建新任务并提交到RLBench任务仓库。

RLBench is a large-scale robotic learning benchmark and learning environment created by the Dyson Robotics Laboratory at Imperial College London. It includes 100 unique manually-designed tasks, ranging from simple target reaching and door opening to complex multi-stage tasks. The dataset provides diverse sensory observations, including RGB, depth data, and segmentation masks, as well as robotic proprioceptive data. Each task is equipped with an unlimited number of demonstrations, which are generated using motion planners based on the waypoints provided during task creation. RLBench aims to accelerate progress in the field of vision-guided manipulation research, particularly for few-shot learning. The dataset is designed with scalability in mind, enabling users to easily create new tasks and submit them to the RLBench task repository.

搜集汇总

背景与挑战

背景概述

RLBench是一个由帝国理工学院戴森机器人实验室开发的大型机器人学习基准和环境，包含100个手工设计的任务，从简单操作到复杂多阶段任务，提供多种感知数据和机器人本体感受信息。它支持无限演示生成，旨在促进视觉引导操作和少样本学习研究，并具有可扩展性，允许用户创建和提交新任务。