cv

ML Software Development Working Student

Infineon Technologies AG

Setup radar DSP preprocessing stack and the entire ML infrastructure from data collection, annotation and loader. Design an end‑to‑end transformer‑based neural network for radar material classification aiming to lawn mower and vacuum cleaner robot. Setup the solution for deploying/compiling our transformer‑based model onto bare‑metal device (PSoC6) using TFLITE/TVM framwork.

• TinyML
• Radar
• Edge Device Deployment

M. Sc in Robotics, Cognition, Intelligence

Technical University of Munich, Bayern, Germany

Artificial Intelligence, Computer Vision

• Machine Learning
• Introduction to Deep Learning
• Introduction to Artificial Intelligence
• Advanced Deep Learning for Robotics
• Robotics
• Robot Motion Planning
• Computer Vision for Multi‑View Geometry
• Vision‑Based Navigation (PR)
• The Evolution of Motion Estimation and Real‑time 3D Reconstruction (SE)

B. Sc in Power Mechanical Engineering

National Tsing Hua University, Hsinchu, Taiwan

Mechanical Engineering

VXP: Voxel-Cross-Pixel Large-scale Image-LiDAR Place Recognition

arXiv

We propose a novel Voxel-Cross-Pixel (VXP) approach, which establishes voxel and pixel correspondences in a self-supervised manner and brings them into a shared feature space. We achieve state-of-the-art performance in cross-modal retrieval on the Oxford RobotCar, ViViD++ datasets and KITTI benchmark, while maintaining high uni-modal global localization accuracy.

Visual-SLAM: Loop Closure and Relocalization

Extend the Visual Odometry framework into Visual SLAM system, and achieve an outstanding performance on EuRoC Vicon Room 1 and EuRoC Machine Hall dataset.

• Visual-SLAM
• Bundle Adjustment

Visual-Inertial Tracking using Preintegrated Factors

Extend the Visual Odometry framework into Visual‑Inertial Odometry. Instead of performing local bundle adjustment by minimizing the reprojection errors in the sliding window of keyframes, we compute preintegrated factor from the IMU measurement to serve as a constraint in the loss function to make the optimization more robust.

• Visual-Inertial Odometry
• Bundle Adjustment

Graph Attention Network for Social Navigation (GAT4SN)

Come up with a GNN‑inspired model named GAT4SN (Graph Attention Network for Social Navigation).

• Deep V‑Learning
• OpenAI Gym
• GNN
• PPO
• DDPG
• DDQN