I work at DoorDash Labs, the robotic and automation arm of DoorDash. Currently focusing on perception and prediction.

Led the object detection team in perception, which consisted of 10+ engineers and two line managers.

The team is responsible for the on-vehicle 3D object detection system, machine learning model training, optimization and deployment, as well as data preprocessing and model evaluation.

Main focus areas include LiDAR-based and camera-based detection, sensor fusion, latency optimization and model training automation. My team also worked closely with partner teams to resolve common on-road perception disengagements and improve the system-level driving performance.

Tech lead manager in Perception-Prediction at Uber ATG. Led a team of 5 based in the SF office. Developed, deployed and improved key perception and prediction deep learning based models on ATG's autonomous vehicles.

Major past projects include

The above projects also led to multiple patents and conference publications (ICRA, IV, WACV, KDD, etc.)

Developed a bottom-up simulator to model the real-time bidding of online ad, by analyzing the massive bidding data. During this project, I developed a modified logistic regression model that can handle periodic input features (e.g. time of day). The simulator was used to quickly evaluate new bidding strategies before A/B tests, and to project the budget spending accurately. The click-through rate model I revised led to a 10% revenue lift during an A/B test that I co-run.