CREPES-X: Hierarchical Bearing-Distance-Inertial Direct Cooperative Relative Pose Estimation System CREPES-X：多层次方位—测距—惯性融合的多机器人直接相对位姿估计系统

Zhehan Li, Zheng Wang, Jiadong Lu, Qi Liu, Zhiren Xun, Yue Wang, Fei Gao, Chao Xu, Yanjun Cao
Zhejiang University · Under review at IEEE Transactions on Robotics 浙江大学 · IEEE Transactions on Robotics 在投（二审）
First author 第一作者
arXiv

Compact bearing–range–inertial hardware, paired with fast and robust direct estimation. 紧凑型方位—测距—惯性感知硬件，结合快速、稳健的直接式估计。

Overview 项目概述

CREPES-X provides direct 6-DoF relative localization without global poses or a shared map. Its custom module integrates an infrared fisheye camera, UWB and an IMU in a cube measuring less than 6 cm on each side. Coded infrared LEDs identify each observed robot, so visual bearings can be associated with range measurements; compared with the previous generation, the module is 77% smaller and 50% lighter. A hierarchical estimator first computes an immediate single-frame pose and then refines it in a multi-frame optimization that tightly couples inter-robot measurements with IMU pre-integration. An angular-consistency test rejects corrupted bearings with 96.8% precision and 94.8% recall when 90% of the inputs are outliers. Real-world experiments report 0.073 m position RMSE and 1.8° rotation RMSE, while the system retains real-time operation with 20 devices.

CREPES-X 无需全局位姿或共享地图，即可直接估计多机器人 6-DoF 相对位姿。自研硬件在边长不超过 6 cm 的模块内集成红外鱼眼相机、UWB 与 IMU；编码式红外 LED 为被观测机器人提供身份信息，使视觉方位与测距数据能够正确关联。与前代相比，模块体积缩小 77%、重量降低 50%。多层次估计器先由单帧观测快速给出位姿，再通过紧耦合机间观测与 IMU 预积分的多帧优化进一步细化。基于角度一致性的方位外点检验在输入外点率达到 90% 时仍有 96.8% 精确率和 94.8% 召回率。真机实验报告的位置 RMSE 为 0.073 m、旋转 RMSE 为 1.8°，并支持 20 台设备实时运行。

Key Contributions 主要贡献

  • Integrated sensing hardware. Infrared fisheye, UWB and IMU sensing are packaged with LED detection, tracking and identity coding in a compact module.
  • Hierarchical direct estimation. A closed-form single-frame estimate provides low latency; tightly coupled multi-frame refinement adds accuracy and robustness without a global frame.
  • Outlier rejection by angular consistency. The proposed test reaches 96.8% precision and 94.8% recall with a 90% bearing-outlier ratio.
  • Real-world scale. The system reports 0.073 m / 1.8° RMSE and remains real time with 20 devices.
  • 一体化感知硬件。在紧凑模块中集成红外鱼眼、UWB 与 IMU，并实现红外 LED 检测、跟踪和身份编码。
  • 多层次直接式估计。单帧闭式估计保证低延迟，紧耦合多帧优化进一步提升精度与鲁棒性，全程不依赖全局坐标系。
  • 基于角度一致性的外点剔除。方位外点率达到 90% 时，所提方法仍取得 96.8% 精确率和 94.8% 召回率。
  • 真机规模验证。系统报告 0.073 m / 1.8° RMSE，并支持 20 台设备实时运行。

BibTeX

@article{li2025crepesx,
  title   = {CREPES-X: Hierarchical Bearing-Distance-Inertial Direct Cooperative Relative Pose Estimation System},
  author  = {Li, Zhehan and Wang, Zheng and Lu, Jiadong and Liu, Qi and Xun, Zhiren and Wang, Yue and Gao, Fei and Xu, Chao and Cao, Yanjun},
  journal = {arXiv preprint arXiv:2512.24688},
  year    = {2025}
}