多机器人协同导航任务训练数据集

该数据集面向三类多机器人协同导航任务，每类任务对应一个文件夹，并在文件夹命名里记录每个任务中智能体和目标点数目的具体设置。在每个文件夹下，分别包含一个“模型文件”子文件夹，记录由pytorch深度学习框架自动保存的checkpoint、metadata等神经网络模型结构和权重文件；另外还包含一个“轨迹训练数据”txt文件，分别对应在该任务下的神经网络训练模型和生成该模型所采用的轨迹数据集，每一条数据里依次包括当前的执行步骤数、智能体采取的动作向量、执行动作后的下一步状态、环境反馈的奖励得分、是否中止、是否中断、以及其他信息。

This dataset is designed for three categories of multi-robot collaborative navigation tasks. Each category corresponds to a dedicated folder, and the specific settings of the number of agents and target points in each task are recorded in the folder's naming convention. Under each folder, there is a subfolder named "model files", which stores neural network model structure and weight files such as checkpoints and metadata automatically saved by the PyTorch deep learning framework. Additionally, there is a "trajectory training data" TXT file, which corresponds to the neural network training model for this task and the trajectory dataset used to generate this model. Each data entry sequentially includes the current execution step count, the action vector taken by the agent, the next state after executing the action, the reward score fed back by the environment, whether the episode is terminated, whether the trajectory is interrupted, and other relevant information.

搜集汇总

背景与挑战

背景概述

该数据集专注于多机器人协同导航任务的训练，涵盖三类任务，每个任务文件夹包含PyTorch框架保存的神经网络模型文件以及记录动作、状态和奖励的轨迹数据。数据集由国防科技大学发布，数据量为97.52MB，适用于研究多机器人系统的协同导航与任务分配。