About Me

I am a 5th-year Ph.D. Candidate at the Laboratory for Information and Decision Systems (LIDS) at MIT, working on robotics and computer vision with Prof. Sertac Karaman.

I obtained my master's degree in 2015 from MIT, and my bachelor's degree in computer engineering in 2013 from HKUST. I was also an intern at Dajiang Innovations (DJI) in 2013. I was the Co-President of MIT-CHIEF (MIT - CHina Innovation and Entrepreneurship Forum) in 2016.

Research Interests

My research interests lie broadly in 3D perception, SLAM, machine learning, compressive sensing, and motion planning, with applications to mobile robots such as driverless cars and drones.

My current focus is on depth image prediction and reconstruction. My goal is to enable onboard perception and navigation of miniature robots (e.g., nano-sized drones) with limited sensing capabilities.

Research Projects

Sparse-to-Dense: Depth Prediction from Sparse Depth Samples and a Single Image

We consider the problem of dense depth prediction from a sparse set of depth measurements and a single RGB image. Since depth estimation from monocular images alone is inherently ambiguous and unreliable, we introduce additional sparse depth samples, which are either collected from a low-resolution depth sensor or computed from SLAM, to attain a higher level of robustness and accuracy. We propose the use of a single regression network to learn directly from the RGB-D raw data, and explore the impact of number of depth samples on prediction accuracy. Our experiments show that, as compared to using only RGB images, the addition of 100 spatially random depth samples reduces the prediction root-mean-square error by half in the NYU-Depth-V2 indoor dataset. It also boosts the percentage of reliable prediction from 59% to 92% on the more challenging KITTI driving dataset. We demonstrate two applications of the proposed algorithm: serving as a plug-in module in SLAM to convert sparse maps to dense maps, and creating much denser point clouds from low-resolution LiDARs.
[publications: C4] [github code] [YouTube video]

Sparse Depth Sensing for Resource-Constrained Robots

In this project we want to enable the onboard sensing and navigational capabilities of nano-sized drones with compact depth sensor that returns very limited amount of data. We address the question: is it possible to reconstruct the geometry of an unknown environment using sparse and incomplete depth measurements? By leveraging depth regularity in structured environments and the compressed sensing framework, we develop practical algorithms for depth image reconstruction, and provide a set of formal results that ascertain the exactness and stability of our approach.
[publications: J2 and C3] [github code] [YouTube video]

On Sensing, Agility, and Computation Requirements in a Stochastic Reward Field

We consider a robotic vehicle tasked with gathering information by visiting a set of spatially-distributed data sources, the locations of which are not known a priori, but are discovered on the fly. We assume a first-order robot dynamics involving drift and that the locations of the data sources are Poisson-distributed. In this setting, we characterize the performance of the robot in terms of its sensing, agility, and computation capabilities. More specifically, the robot's performance is characterized in terms of its ability to sense the target locations from a distance, to maneuver quickly, and to perform computations for inference and planning. We also characterize the performance of the robot in terms of the amount and distribution of information that can be acquired at each data source.
[publications: J1 and C2]

Visual-Inertial Velocity Estimation

To provide a robust and accurate velocity estimation for micro aerial vehicles, we use a monocular camera and an inertial measurement unit. We proposed a coarse-to-fine algorithm based on multiple feature correspondences across three consecutive image frames.
[publications: C1]

Publications & Preprints