LoHoRavens

LoHoRavens是由慕尼黑工业大学创建的一个模拟基准数据集，专注于长期语言条件下的机器人桌面操作任务。该数据集包含10个长期语言条件任务，分为可见任务和不可见任务，用于评估机器人的泛化性能。数据集的创建旨在解决机器人如何在没有昂贵注释演示的情况下执行长期任务的问题，特别是在需要颜色、大小、空间、算术和引用等多种推理能力的场景中。通过模拟环境生成专家演示，数据集支持机器人学习复杂的操作技能，并测试其在不同任务中的表现。

LoHoRavens is a simulated benchmark dataset developed by the Technical University of Munich, focusing on robotic tabletop manipulation tasks conditioned on long-horizon language instructions. This dataset includes 10 long-horizon language-conditioned tasks, which are categorized into visible tasks and invisible tasks for evaluating robotic generalization performance. The dataset is designed to address the challenge of enabling robots to execute long-horizon tasks without costly annotated demonstrations, particularly in scenarios requiring multiple reasoning capabilities such as color, size, spatial, arithmetic, and referential reasoning. Expert demonstrations are generated through simulated environments, and the dataset supports robotic learning of complex manipulation skills as well as testing robots' performance across different tasks.

搜集汇总

背景与挑战

背景概述

LoHoRavens是由慕尼黑工业大学创建的模拟基准数据集，专注于长期语言条件下的机器人桌面操作任务，包含10个任务以评估机器人的泛化性能。它旨在解决机器人执行长期任务时无需昂贵注释演示的问题，支持在涉及多种推理能力的场景中学习复杂操作技能。