Aaron Low Weng Soon
SUMMARY
Senior robotics engineer with six years in autonomous vehicles, specialising in multi-sensor calibration and fusion. During my career, I've worked on systems that utilise a combination of deep learning and classical methods like factor graphs, multi-view geometry, and SLAM. I work end to end, developing the algorithms and building the scalable, automated tooling and infrastructure that makes them fast to iterate on and reliable to run.
WORK EXPERIENCE
- • Designed and implemented multi-sensor calibration algorithms (camera, IMU, lidar) combining deep learning with multi-view geometry and SLAM.
- • Built scalable automated pipelines running calibration continuously supporting a fleet of vehicles spread across multiple locations globally.
- • Our improvements led to a 4x speedup in turnaround time for calibrating a vehicle.
- • Built an end-to-end deep learning pipeline for training and evaluating visual feature extraction and calibration models, covering dataset generation, curation, and loading.
- • Implemented metrics and automated validation that detect miscalibration and drift, catching issues before they reached downstream users.
- • Developed internal tooling to deploy, track, and maintain per-vehicle calibrations across the fleet.
- • Led technical direction, mentored junior engineers, and helped shape hiring decisions
- • Developed solutions for sensor (cameras, LIDARs, radars, IMU) calibration and validation of autonomous vehicles.
- • Researched deep learning solutions for sensor calibration as well as implementing the infrastructure for training and evaluation.
- • Built and maintained cloud-based systems for continuous validation of vehicle sensor data.
- • Our automated system ran daily on a fleet of hundreds of vehicles allowing data-driven calibrations.
- • Automated dataset generation and curation for algorithm research.
- • Worked on vehicle hardware to deploy online sensor calibration algorithms.
- • Researched deep learning for camera-to-LiDAR calibration and deployed a RegNet based model into production with TensorRT.
PUBLICATIONS
- • UniCal: a Single-Branch Transformer-Based Model for Camera-to-LiDAR Calibration and Validation — a Transformer architecture using self-attention to perform camera-to-LiDAR calibration and validation in a single branch.
- • Depth to Colour Translation for 3D Hand Pose Estimation From Monocular RGB With Generative Adversarial Networks — MEng thesis, Imperial College London.
EDUCATION
- • Research area: 3D Pose Estimation
- • First Class Honours • Dean's List (top 10% of class) Year 2
TEACHING
PROJECTS
Robot Navigation- • Autonomous robot that manoeuvres a building and takes an elevator to obtain and deliver snacks.
- • Prize Winner (“Best use of Amazon Web Services”) Parking space detection web service using machine learning to detect cars in parking lots.
- • Featured by GitHub in "Top entries from Game Off 2021". A 2D platformer written in Unity with C#.
TECHNICAL SKILLS
Programming: Python • C++ • C#
ML and Robotics: PyTorch • TensorFlow • ROS • TensorRT • GTSAM
Infrastructure: AWS • Azure • Kubernetes • Spark • Terraform • Docker • Jenkins • Serverless • Bazel • CMake
Tools and Data: Git • Linux • Bash • Pandas • SQL • Unity
Languages: English (Native) • Malay (Limited Working)