Menghao Wu

Software & AI Engineer with 10+ years of experience spanning robotics perception, deep learning model design, training, and GPU-based optimization. Expert in the end-to-end ML lifecycle — including model design, training, profiling, optimization, benchmarking, and deployment. Proven ability to enhance model performance through advanced training strategies, acceleration techniques, and inference pipeline optimization. Experienced in MLOps integration, continuous benchmarking, and scalable deployment for real-time intelligent systems. Passionate about optimizing model training, resource efficiency, and inference performance across modern compute platforms.

SKILLS

Machine Learning & Optimization:
Model design, training, benchmarking, end-to-end ML lifecycle management, inference acceleration (TensorRT, ONNX, custom CUDA kernels), CUDA programming and GPU profiling, performance tuning and real-time deployment

Programming Languages:
Python, C++, CUDA, Bash, Linux

Frameworks & Tools:
PyTorch, TensorFlow, OpenCV, scikit-learn, Pandas, Git, Jupyter

MLOps & DevOps:
MLflow, Docker, Kubernetes, Docker, AWS, CI/CD automation, shell scripting

Collaboration & Documentation:
Agile/Scrum, cross-functional collaboration, technical writing and research documentation

PUBLICATIONS

Wu, Menghao, et al. "The multi-dimensional actions control approach for obstacle avoidance based on reinforcement learning." Symmetry 13.8 (2021): 1335.
Du, Shitong, Wu, Menghao et al. "LiDAR odometry and mapping based on semantic information for outdoor environment." Remote Sensing 13.15 (2021): 2864.
Wu, Menghao, et al. "The actor-dueling-critic method for reinforcement learning." Sensors 19.7 (2019): 1547.
Wu, Menghao, et al. "Graph signal sampling with deep Q-learning." 2020 International Conference on Computer Information and Big Data Applications (CIBDA). IEEE, 2020

EXPERIENCE

GIM Robotics, Finland, Senior Robotics Engineer
JUL 2020 - PRESENT

• Designed, trained, and optimized PointPillars-based 3D object detection models using MMDetection3D and NVIDIA TAO Toolkit, achieving >95% accuracy across diverse industrial and outdoor scenarios.
• Led the full ML lifecycle for object detection — data collection, annotation, training design, and validation — with large-scale AWS multi-GPU acceleration for faster experimentation and hyperparameter tuning.
• Deployed optimized inference pipelines on Jetson and x86 GPU platforms, applying CUDA profiling, latency optimization, and TensorRT acceleration for real-time performance.
• Built an automated ML pipeline on AWS supporting data ingestion, model training, benchmarking, and deployment across distributed test fleets.
• Developed and benchmarked YOLO-based detectors trained on Finnish winter datasets, enhancing perception robustness under snow, low-light, and glare conditions.
• Collaborated cross-functionally with hardware and perception teams to integrate optimized ML modules into production-grade robotic perception stacks for ground segmentation, obstacle detection, and collision awareness.

Aalto University, Finland, Research Assistant
OCT 2016 - MAY 2018

• Conducted research on reinforcement learning and deep neural network optimization within the Machine Learning for Big Data group at Aalto University’s Department of Computer Science.
• Designed and trained machine learning and reinforcement learning models for object detection and continuous control, leveraging TensorFlow, PyTorch, and Keras for model development and experimentation.
• Implemented and validated autonomous obstacle avoidance on a TurtleBot platform, integrating RL-based policies with classical control frameworks.
• Utilized Aalto’s large-scale GPU computing servers for high-performance training, hyperparameter tuning, and model benchmarking.
• Contributed to data processing, visualization, and performance analysis pipelines for evaluating model generalization and real-world applicability.

ZTE Corporation, Beijing, China, Software Engineer Intern
AUG 2015 - DEC 2015

• Contributed to software development, testing, and related tasks.

EDUCATION

Harbin Engineering University, China - Ph.D., Master
SEP 2014 - JUL 2021

• Conducted research on deep reinforcement learning, continuous control, and hierarchical policy design for perception systems.
• Developed CNN-based landmark recognition models (e.g., VGG) to enhance mobile robot navigation and localization.
• Gained extensive experience with TensorFlow, PyTorch, and Keras for deep learning model design and experimentation.
• Served as a Visiting Researcher at Aalto University, focusing on reinforcement learning applications in robotics.

Northeastern University, China - Bachelor
SEP 2010 - JUL 2014

• Designed a smart car with C++/ARM microcontroller-based sensor fusion for road following and obstacle avoidance.
• National award winner in embedded programming design competitions.

OTHERS

Languages: English (Fluent), Chinese (Native), Finnish (Basic)

Github: https://github.com/MengWoods

Technical blogs: https://mengwoods.github.io/

YouTube Channel: https://www.youtube.com/@menghao_w

Hobby: Painting, graphic design.