多任务长程任务模仿学习数据集

数据集名称：多任务长程任务模仿学习数据集（编号2020AAA0108902-004）数据集内容：建立了一个全新的多任务长程任务模仿学习数据集，旨在弥补现有数据集的不足并满足实际应用的需求。该数据集采用MuJoCo仿真环境进行采集，同步收集低维观测和执行动作，涵盖多种任务例如开抽屉、抓取物体、堆叠积木等。数据集总计约300条操作轨迹，包括机器人本体感知信息、物体信息和执行动作。该数据集不仅为机器人长程领域提供了宝贵的资源，还为持续学习等新兴研究领域设定了新的测试基准。数据来源：通过仿真分析获得。采集时间及地点：数据采集时间为2023年，地点为中国科学院自动化研究所。本数据集旨在弥补现有数据集的不足并满足实际应用的需求，采用MuJoCo仿真环境进行采集。数据集同步收集了低维观测和执行动作，涵盖多种任务，如开抽屉、抓取物体、堆叠积木等。数据集总计约300条操作轨迹，包括机器人本体感知信息、物体信息和执行动作。数据量为17.8MB。Dataset Name: Multi-task Long-horizon Task Imitation Learning Dataset (No. 2020AAA0108902-004) Dataset Content: A novel multi-task long-horizon task imitation learning dataset is constructed to address the limitations of existing datasets and meet the requirements of practical applications. The dataset is collected using the MuJoCo simulation environment, synchronously capturing low-dimensional observations and executed actions. It covers a variety of tasks including drawer opening, object grasping, block stacking, and more. The dataset contains approximately 300 operational trajectories, which include robot proprioceptive information, object-related information, and executed actions. This dataset not only provides valuable resources for the field of robotic long-horizon tasks, but also establishes a new test benchmark for emerging research areas such as continual learning. Data Source: Obtained via simulation analysis. Data Collection Time and Location: The data was collected in 2023 at the Institute of Automation, Chinese Academy of Sciences. Total Dataset Size: 17.8 MB

搜集汇总

背景与挑战

背景概述

该数据集是一个多任务长程任务模仿学习数据集，旨在弥补现有数据集的不足。它采用MuJoCo仿真环境采集，同步收集低维观测和执行动作，涵盖开抽屉、抓取物体、堆叠积木等多种任务，总计约300条操作轨迹，包含机器人本体感知信息、物体信息和执行动作，数据量为17.8MB。