Large-scale deployments of robot teams are challenged by the need to share high-resolution perceptual information over low-bandwidth communication channels. Individual size, weight, and power constrained robots rely on environment models to assess navigability and safely traverse unstructured and complex environments. State of the art perception frameworks construct these models via multiple disparate pipelines that reuse the same underlying sensor data, which leads to increased computation, redundancy, and complexity. To bridge this gap, this paper introduces GIRA -- an open-source framework for compact, high-resolution environment modeling using Gaussian mixture models (GMMs). GIRA provides fundamental robotics capabilities such as high-fidelity reconstruction, pose estimation, and occupancy modeling in a single continuous representation.

翻译：大规模机器人团队的部署面临一项挑战：需通过低带宽通信信道共享高分辨率感知信息。受尺寸、重量和功耗限制的单个机器人依赖环境模型评估可通行性，并在非结构化复杂环境中安全导航。现有感知框架通过多个独立管道重复使用同一底层传感器数据构建环境模型，导致计算量增加、冗余和复杂性上升。为弥合这一差距，本文提出GIRA——一个基于高斯混合模型（GMM）实现紧凑高分辨率环境建模的开源框架。GIRA在单一连续表示中提供高保真重建、位姿估计和占用建模等基础机器人能力。