Huarui Zhang

A Deep Learning Positioning Framework Fusing Pseudorange, Carrier Phase, and Doppler Observations

IEEE UPINLBS 2025 · First author

这项工作把伪距、载波相位和多普勒观测放入同一定位框架，并使用 PyTorch 与 pyrtklib 完成训练和评估。在约 2.5 km 的数据上，二维和三维平均误差相比 PSR+CP 基线分别下降 59.5% 和 36.6%。

This work combines pseudorange, carrier phase, and Doppler observations in one positioning framework. On a 2.5 km sequence, it reduced mean 2D and 3D errors by 59.5% and 36.6% against the PSR+CP baseline.

TLDR 多普勒观测为长序列定位补充了稳定的速度信息。

基于物理信息神经网络的 GNSS/INS 组合导航

2026 · First author

我把组合导航中的物理约束加入网络训练，并使用 78 维卫星与 IMU 特征预测导航状态。在 GVINS urban driving 数据上，三维 RMSE 从 36.03 m 降至 13.09 m。

I incorporated navigation constraints into network training and used 78-dimensional satellite and IMU features to estimate navigation states. On the GVINS urban driving sequence, 3D RMSE dropped from 36.03 m to 13.09 m.

TLDR 物理约束减少了观测条件变化时的模型偏移。

全国大学生智能汽车竞赛 · 走马观碑

2026

我负责主控板与驱动板、嵌入式程序和整车调试。设计中重新处理了重心、散热和走线，整车实测速度约为 2.7 m/s。

I worked on the main board, driver board, embedded software, and vehicle testing. The car was revised for center of gravity, cooling, and wiring, and reached approximately 2.7 m/s.

TLDR 从板卡设计到整车调试，工作贯穿完整备赛过程。

全国大学生智能汽车竞赛 · 微缩光电

2025

我参与主控、驱动和 SD 模块的原理图与 PCB Layout，并配合完成车模减重和 PWM 控制验证。整车减重约 100 g。

I worked on schematics and PCB layout for the controller, motor driver, and SD module, and helped verify weight reduction and PWM control. The car was lightened by approximately 100 g.

TLDR 减重、电路和控制需要放在整车上一起验证。

中国机器人大赛暨 RoboCup 中国赛

Nov, 2024

我使用 SolidWorks 完成背筐弹射装置的结构设计，以 STM32F103C8T6 为核心完成原理图、PCB Layout 和串口通信程序，并参与机器狗整机调试。

I designed the basket launcher in SolidWorks, built the schematic, PCB layout, and serial communication code around an STM32F103C8T6, and participated in robot testing.

TLDR 结构、电路和控制都围绕实际比赛动作展开。

2026 MCM Problem C

Mathematical Contest in Modeling · Honorable Mention

团队完成了模型建立、数据分析、图表和论文撰写。我负责部分建模与计算，并使用 SUR、USI/Log-USI 和 10,000 次 Monte Carlo 模拟检验结果。

Our team completed the modeling, analysis, figures, and paper. I contributed to the model and computation, including SUR, USI/Log-USI, and 10,000 Monte Carlo runs.