JPL Mars Yard Duo Dataset

我们提出了一种系统解决方案，可以使用热图像和惯性测量为飞行机器人团队实现数据高效，分散的状态估计。每个机器人都可以独立飞行，并在可能的情况下交换数据以完善其状态估计。我们的系统前端应用在线光度校准来优化热图像，从而增强特征跟踪和位置识别。我们的系统后端使用协方差交叉融合策略来忽略代理之间的互相关，从而降低内存使用量和计算成本。通信管道使用本地聚合描述符 (VLAD) 的向量来构建需要低带宽使用的请求-响应策略。我们在合成数据和真实数据上测试我们的协作方法。我们的结果表明，相对于个体代理方法，所提出的方法通过多达46% 条轨迹估计进行了改进，同时减少了多达89% 的通信交换。数据集和代码发布给公众，扩展了已经公开的JPL xVIO库。

We present a systematic solution that enables data-efficient, decentralized state estimation for teams of flying robots using thermal images and inertial measurements. Each robot can operate independently in flight and exchange data when feasible to refine its state estimation. The front-end of our system applies online photometric calibration to optimize thermal images, thereby enhancing feature tracking and place recognition. The back-end adopts a covariance intersection fusion strategy to eliminate the need for explicit modeling of cross-correlations between agents, reducing memory usage and computational cost. The communication pipeline constructs a low-bandwidth request-response strategy by leveraging vectors of locally aggregated descriptors (VLAD). We evaluate our collaborative approach on both synthetic and real-world datasets. Our results demonstrate that the proposed method improves trajectory estimation by up to 46% compared to individual agent-based approaches, while cutting down communication exchanges by up to 89%. All datasets and code are publicly released, extending the previously open-sourced JPL xVIO library.

搜集汇总

背景与挑战

背景概述

该数据集提出了一种基于热图像和惯性测量的系统解决方案，用于实现飞行机器人团队的分散状态估计，通过在线光度校准和协方差交叉融合策略提升估计精度并减少通信成本。它在合成和真实数据上测试，相较于个体方法，轨迹估计改进达46%，通信交换减少达89%，并扩展了公开的JPL xVIO库。数据集和代码已公开发布，支持进一步研究。