Multi-Sensor Robot Navigation
📝 Detailed writeup coming soon — key information is shown below.
Multi-Sensor Robot Navigation Research Research
Overview
This project focused on robot navigation in low-light and visually degraded environments where standard RGB cameras fail. The approach combined stereo thermal imaging, LiDAR, and RGB-D sensors into a unified navigation stack, deployed and validated on a real UGV (Unmanned Ground Vehicle).
Sensor Pipeline & Calibration
- Built a stereo thermal / RGB-D / LiDAR multi-sensor data acquisition pipeline
- Implemented hardware-level time synchronization across heterogeneous sensors
- Completed intrinsic and extrinsic calibration for all sensor pairs
Navigation Stack on AGX Orin
- Integrated BridgeDepth (thermal depth estimation), NVBlox (neural volumetric mapping), Nav2, and SLAM pose input on NVIDIA AGX Orin
- Validated point-goal navigation on a real UGV platform in indoor environments
4D Radar Odometry
- Connected RPM-Net (scan-to-scan soft correspondences) and Radar-Transformer (scan-to-submap hard correspondences) into a radar-inertial odometry backend (IEKF / VGICP)
- Evaluated on ColoRadar and SNAIL benchmark datasets
- Related work accepted as ISPRS 2026 Extended Abstract
Figures, videos, and quantitative results will be added here.
📝 详细内容即将补充，以下为主要信息。
多传感器机器人导航与定位研究 科研
项目概述
本项目面向低照度与视觉退化环境中的机器人导航问题。在标准 RGB 相机失效的场景下，通过融合双目热红外成像、LiDAR 与 RGB-D 传感器，构建完整导航栈，并在真实 UGV（无人地面车辆）上完成实机部署与验证。
传感器链路与标定
- 搭建双目热红外 / RGB-D / LiDAR 多传感器采集链路
- 实现跨异构传感器的硬件级时间同步
- 完成所有传感器对的内外参标定
AGX Orin 上的导航栈
- 在 NVIDIA AGX Orin 上集成 BridgeDepth（热红外深度估计）、NVBlox（神经体积建图）、Nav2 与 SLAM 位姿输入
- 在室内环境下的真实 UGV 平台上完成 point-goal 导航实机验证
4D 雷达里程计
- 将 RPM-Net（scan-to-scan 软对应）与 Radar-Transformer（scan-to-submap 硬对应）接入雷达惯性里程计后端（IEKF / VGICP）
- 在 ColoRadar 与 SNAIL 基准数据集上完成评测
- 相关工作已被接受为 ISPRS 2026 Extended Abstract
图片、视频与定量结果将在后续补充。