唐隆斌

工作经历

感知算法工程师
小马智行 Pony.ai 2025.7 - 至今
  • 参与 RoboTruck 多模态 BEV 感知基础模型研发,设计 latent-to-early BEV temporal fusion:将历史帧 BEV feature 对齐并残差融合到当前帧,替代高成本多帧点云拼接;优化 OHEM 类别采样,部分 AP 提升 10-30%,漏检/误检降低约 30-40%。
  • 负责 RoboTruck trailer detection 优化,基于 LiDAR BEV regression 引入运动学先验、动态 ROI 与 Kalman Filter 后处理,并通过数据重采样、onboard C++ 部署和闭环验证,使整体角度误差降低约 12%。
  • 优化点云障碍物合并模型,结合 PointNet/Transformer 与上游语义先验判断 polygon 合并关系,缓解目标过切分,提升 tracking/PNC 输入完整性。
  • 重构多源 heuristic hint selection,统一 box、track 与 FCN polygon 表示选择,基于几何一致性和 ROI/车道关系,使 under/over-segmentation 问题整体降低约 10%。

技术:BEV Perception,LiDAR-Camera Fusion,Transformer,C++/Python

机器人算法实习生
星海图 GalaXeaAI 2024.2 - 2024.6
  • 参与移动机器人 3D Semantic Occupancy 数据闭环,基于环视相机与 LiDAR SLAM 地图生成语义 OCC 监督,并完成数据清洗、点云修正与训练实验分析。
  • 构建 2D-to-3D 自动标注流程,利用 SAM 生成语义 mask,经 IPM 投影与 OCC 连通块传播 3D 语义;开发 BEV 标注工具补充门框等困难类别。
  • 参与移动机器人感知-规划-控制联调,部署 DL planner,调试四舵轮底盘运动学/控制,并基于 ROS2 与 VR 手柄实现机械臂 6D 遥操作原型。

技术:3D OCC,Auto Labeling,LiDAR SLAM,DL Planner,ROS2/Kinematics

项目和研究经历

Robotics Planning and Learning Research
新加坡国立大学 | CORE Lab | Lin Zhao 2024.9 - 2025.6
  • Learning Agile Gate Traversal:参与 NN 自适应调参的可微 MPC 控制框架,构建 Isaac Sim/Lab 窄门与动态扰动实验,产出 IROS 2026 论文。
  • AERO-MPPI:参与 LiDAR anchor-guided ensemble MPPI 规划控制,搭建 ODE 求解器、无人机动力学与仿真评估环境,产出 ICRA 2026 论文。
  • MacroNav:参与轻量上下文编码器与 graph-RL 导航策略设计,推进未知环境真实部署与实验分析,产出 RA-L 2026 论文。
Autonomous Vehicle for mail and parcel delivery
新加坡国立大学(苏州)研究院 | 西交利物浦大学 | Jie Sun 2023.10 - 2024.6
从零搭建园区无人配送车系统,负责六轮差速底盘设计、装配与驱动开发,接入 Autoware.universe 多传感器链路;实现 SLAM 建图定位、点云分割、路径规划与底盘控制,产出 IRAC 2024 论文及发明专利。
HITSZ VEX 机器人队队长
哈尔滨工业大学(深圳)| VEXU Robotics Team 2022.1 - 2023.6
从零组建并负责校 VEX 机器人队,兼任机械设计与程序开发;完成底盘运动学建模、系统辨识、LQR 控制、路径规划与遥控映射,带队获得区域赛前三、国赛第二、VEXU 亚洲冠军、世界赛八强等成绩。
论文发表
  1. Kuankuan Sima, Longbin Tang, Zhenyu Yang, Haozhe Ma, Lin Zhao. MacroNav: Multi-Task Context Representation Learning Enables Efficient Navigation in Unknown Environments. RA-L 2026.
  2. Xin Chen, Rui Huang, Longbin Tang, Lin Zhao. AERO-MPPI: Anchor-Guided Ensemble Trajectory Optimization for Agile Mapless Drone Navigation. ICRA 2026.
  3. Tianchen Sun, Bingheng Wang, Nuthasith Gerdpratoom, Longbin Tang, Yichao Gao, Lin Zhao. Learning Agile Gate Traversal via Analytical Optimal Policy Gradient. IROS 2026.
  4. Runyu Lei, Longbin Tang, Jiaxin Guo, Jie Sun, Qinglei Bu. Autonomous Delivery Vehicle System Based on Multi-Sensor Data Fusion. IRAC 2024.
专利成果
  1. 唐隆斌, 王锐, 何铭, 刘能峰, 一种差分舵轮及小车, 公开号 CN117104334A
  2. 唐隆斌, 郭佳鑫, 雷润宇, 卜庆雷, 孙捷, 一种无人驾驶车辆的车轮控制方法以及装置

Longbin Tang

Work Experience

Perception Algorithm Engineer
Pony.ai 2025.7 - Now
  • Contributed to a RoboTruck multi-modal BEV perception base model and designed latent-to-early BEV temporal fusion: aligning and residually fusing historical BEV features into current-frame features to replace expensive multi-frame point-cloud concatenation; optimized OHEM sampling, improving partial AP by 10-30% and reducing missing/false detections by ~30-40%.
  • Owned RoboTruck trailer detection optimization with LiDAR BEV regression, kinematic priors, dynamic ROI, and Kalman Filter post-processing; improved robustness through data resampling, onboard C++ deployment, and validation loops, reducing overall angle error by ~12%.
  • Improved point-cloud obstacle merging by combining PointNet/Transformer with upstream semantic priors to infer polygon merge relationships, mitigating over-splitting and improving tracking/PNC input completeness.
  • Refactored multi-source heuristic hint selection across box, track, and FCN polygon representations, using geometry consistency and ROI/lane relations to reduce under/over-segmentation issues by ~10%.

Techs: BEV Perception, LiDAR-Camera Fusion, Transformer, C++/Python

Robotics Algorithm Intern
GalaXeaAI 2024.2 - 2024.6
  • Contributed to the mobile robot 3D Semantic Occupancy data loop by generating semantic OCC supervision from surround-view cameras and LiDAR SLAM maps, covering data cleaning, point-cloud correction, and training analysis.
  • Built a 2D-to-3D auto-labeling workflow using SAM masks, IPM projection, and OCC connected components to propagate 3D semantics; developed a BEV annotation tool for hard classes such as door frames.
  • Supported mobile robot perception-planning-control integration by deploying a DL planner, tuning four-steering-wheel chassis kinematics/control, and prototyping ROS2-based 6D arm teleoperation with VR controllers.

Techs: 3D OCC, Auto Labeling, LiDAR SLAM, DL Planner, ROS2/Kinematics

Projects & Research

Robotics Planning and Learning Research
National University of Singapore | CORE Lab | Lin Zhao 2024.9 - 2025.6
  • Learning Agile Gate Traversal: contributed to an NN-adaptive differentiable MPC control framework and Isaac Sim/Lab gate-traversal/disturbance experiments, resulting in an IROS 2026 paper.
  • AERO-MPPI: contributed to LiDAR anchor-guided ensemble MPPI planning/control, ODE solver, UAV dynamics, and simulation evaluation, resulting in an ICRA 2026 paper.
  • MacroNav: worked on lightweight context encoding and graph-RL navigation policy design, real-world deployment validation, and experiment analysis, resulting in a RA-L 2026 paper.
Autonomous Vehicle for Mail and Parcel Delivery
NUSRI | XJTLU | Jie Sun 2023.10 - 2024.6
Built an autonomous campus delivery vehicle from scratch, covering six-wheel differential chassis design, assembly, driver development, and Autoware.universe multi-sensor integration; implemented SLAM, point-cloud segmentation, planning, and chassis control, leading to an IRAC 2024 paper and related patent.
HITSZ VEX Robotics Team Captain
Harbin Institute of Technology (Shenzhen) | VEXU Robotics Team 2022.1 - 2023.6
Founded and led the HITSZ VEX robotics team while owning mechanical design and software development; implemented chassis kinematics, system identification, LQR control, path planning, and teleop mapping, leading the team to regional top-3, national runner-up, VEXU Asia Champion, and World Championship top-8 results.
Publications
  1. Kuankuan Sima, Longbin Tang, Zhenyu Yang, Haozhe Ma, Lin Zhao. MacroNav: Multi-Task Context Representation Learning Enables Efficient Navigation in Unknown Environments. RA-L 2026.
  2. Xin Chen, Rui Huang, Longbin Tang, Lin Zhao. AERO-MPPI: Anchor-Guided Ensemble Trajectory Optimization for Agile Mapless Drone Navigation. ICRA 2026.
  3. Tianchen Sun, Bingheng Wang, Nuthasith Gerdpratoom, Longbin Tang, Yichao Gao, Lin Zhao. Learning Agile Gate Traversal via Analytical Optimal Policy Gradient. IROS 2026.
  4. Runyu Lei, Longbin Tang, Jiaxin Guo, Jie Sun, Qinglei Bu. Autonomous Delivery Vehicle System Based on Multi-Sensor Data Fusion. IRAC 2024.
Patents
  1. Tang Longbin, Wang Rui, He Ming, Liu Nengfeng, A Differential Steering Wheel and Vehicle, Publication No. CN117104334A
  2. Tang Longbin, Guo Jiaxin, Lei Runyu, Bu Qinglei, Sun Jie, A Wheel Control Method and Device for Autonomous Vehicles