About

The short history of my life is: I was born in Singapore, then moved to Vancouver, BC, Canada at the age of 11, then I moved to the US for grad school (Pittsburgh, PA) and work (San Francisco bay area, CA). My parents are both Shanghainese; they moved from Shanghai to the US for grad school, then to Singapore for work, and now they reside in Canada.

1 Work

I work in what’s called Spatial AI nowadays. Basically, I have a decade of experience in making sense of the 3D world via simultaneous localization and mapping (SLAM) with lidar, cameras, and other sensors, and more recently, stuff like Gaussian splatting. I currently work at NVIDIA on their self driving car project, on context fusion.

resume

1.1 NVIDIA

NVIDIA is working on Level 2 automated driver assistance systems with customers such as Mercedes Benz. I’m on the context fusion team and working on static world perception (road topology and geometry).

Sorry, no pictures to share here.

1.2 Main Street Autonomy

Main Street Autonomy is a startup making localization, mapping, and calibration solutions using cameras, lidar, depth cameras, and other stuff.

I implemented:

Screenshot extracted from the background video of mainstreetautonomy.com.

FIGURE 1 Screenshot extracted from the background video of mainstreetautonomy.com.

Download

1.3 Tesla

Worked on autolabeling and 3D reconstruction.

For autolabeling, we combined trajectories from multiple cars using visual odometry and bundle adjustment, and swept out their camera-space segmentation masks onto a road surface (and optimized said road surface to that everything lines up). I owned most of the pipeline as well as the ground surface reconstruction and sweeping logic.

Autolabeling. Screenshot from Tesla AI Day 2022.

FIGURE 2 Autolabeling. Screenshot from Tesla AI Day 2022.

Download

For 3D reconstruction, I implemented the renderer from scratch, but we used an off-the-shelf algorithm for the actual reconstruction. The renderer is technically capable of rendering Gaussian splats, but we hadn’t been able to ship the actual Gaussian splatting by CVPR 2023. They were later able to ship Gaussian splatting by CVPR 2025.

3D reconstruction demo. Screenshot from Ashok's CVPR WAD Keynote 2023.

FIGURE 3 3D reconstruction demo. Screenshot from Ashok’s CVPR WAD Keynote 2023.

Download

I also worked on:

1.4 Tempus Ex Machina

Worked on sports field registration and human pose estimation. Sorry, no public visualizations.

1.5 Ouster

At Ouster, I was the first software engineer. I grew the team from 1 to like 10ish and implemented amazing 6 DoF, real time, continuous time SLAM!

Lidar point cloud.

FIGURE 4 Lidar point cloud.

Download
Aligning open-loop trajectory to GPS.

FIGURE 5 Aligning open-loop trajectory to GPS.

Download
With GPS but no loop closure alignments.

FIGURE 6 With GPS but no loop closure alignments.

Download
With loop closure.

FIGURE 7 With loop closure.

Download

2 Education

At CMU, my advisor was George Kantor. I worked on SLAM using a combination of cameras and a lidar, for an autonomous off-road vehicle.

My thesis is called Vision-Enhanced Lidar Odometry and Mapping. It is about combining several different SLAM algorithms, each with different strengths and weaknesses, so that a robot can function robustly in many different environments.

3 Contact

My email address is the same as my full name except with an ’@’ in between my first and middle names and a dot between my middle and last name.

email

On many websites I have the username dllu except some websites where I was too late and the four letter username was already taken.

websitelink
githubgithub.com/dllu
xx.com/dllu
wikipediaen.wikipedia.org/wiki/User:dllu
hacker newsnews.ycombinator.com/user?id=dllu
redditreddit.com/u/dllu
Table 1 Links to my account on various websites.

4 Publications

4.1 Journal papers

4.2 Preprints

4.3 Technical reports