Chris Wei Xun Lai

About Chris Wei Xun Lai

Graduated undergraduate at University of California Berkeley in Fall 2023 with a degree in Electrical Engineering and Computer Sciences with pending M.S. Applications. As a Lead Perception Researcher at AI Racing Tech, I apply my computer vision and robotics skills to develop and implement the perception stack of the autonomous racing cars that compete in the Indy Autonomous Challenge. I have over a year of experience in this role, where I have achieved several milestones, such as: Representing the top US Indy Autonomous Racing Team in national and international events, showcasing our innovative solutions and achievements. Pioneering the development of the Lidar-Camera racecar detection pipeline that utilizes state-of-the-art algorithms and techniques, such as YOLOv7, Euclidean Clustering, and Camera Projection. Setting up PTP on the racecar perception hardware package, which consists of multiple sensors and devices that are connected to a Cisco Series switch, ensuring high-performance and reliability. I am also a Student Researcher at UC Berkeley Electrical Engineering & Computer Sciences (EECS) and FHL Vive Center for Enhanced Reality, where I work on various projects related to biomimetic milli systems and enhanced reality. I am currently investigating the suitability of DQR on constrained RL tasks from OpenAI\'s safety gym simulation with Aaron Kandel. I am passionate about advancing the field of autonomous systems and perception, and I strive to learn new skills and technologies that can help me achieve this goal. I hold a BS in EECS from UC Berkeley, and I have completed transfer credits.

Experience

1. Graduate Researcher

   Uc Berkeley Electrical Engineering & Computer Sciences Eecs

   Nov 2023 — Present · Berkeley, CA, US

   Leadership & Strategy: Directed the team\'s efforts in model training, data collection, and the intricate process of single and multi-modal sensor calibration. My role was pivotal in ensuring the hardware and software configurations were optimized for high-stakes racing environments, emphasizing safety and reliability.Innovative Solutions: Successfully deployed a cutting-edge sensor-fusion perception system, designed for high-speed racing up to 200 mph. This system set new standards in 3-D object detection, achieving a groundbreaking mean average precision (mAP50-95) of 96.7%, while maintaining an inference throughput of 200Hz, powered by Extended Kalman Filter tracking.Data Annotation System: Conceived and executed a state-of-the-art data annotation system, handling over 3 million raw 1080p footage from autonomous vehicles. This system streamlined the labeling process through unsupervised and semi-supervised methods, significantly reducing the need for human verification.Tool Development: Developed a SAM-based image segmentation labeling tool, enhancing the efficiency of the human verification process. This innovation enabled rapid labeling of images per hour, integrating advanced features like label tracking using De-AOT and supporting user concurrency with technologies such as Celery and RabbitMQ.Technology Optimization: Fine-tuned Yolov8-nano for superior image segmentation-based detection, incorporating lidar and radar for sensor fusion. This involved meticulous hyperparameter optimization, culminating in the deployment of the solution on an ADAS-grade computer, leveraging C++ & TensorRT for unmatched performance.

Education

• University of California, Berkeley

  Bachelor of Science - BS, Electrical Engineering and Computer Science

  2020 — 2023

• University of California, Berkeley

  Master of Science - MS, Electrical Engineering and Computer Sciences