I am a third-year PhD student in Computer Science at Australian National University where I am advised by Miaomiao Liu, David Ahmedt-Aristizabal and Chuong Nguyen. Before that, I received my Master of Philosophy degree supervised by Liang Zheng from Australian National University in 2023. I am currently a research intern at X-Humanoid, working closely with Jony Zhang. My research interests include control & planning in robotics, reinforcement learning, vision perception, and 3D reconstruction.
FlowOp uses sceneflow-conditioned diffusion to retarget arbitrary animation character motions to humanoid robots without body correspondence.
RobotPan predicts metric-scaled compact 3D Gaussians from sparse surround-view inputs for real-time 360° rendering and reconstruction on humanoid robots.
Heracles is a state-conditioned diffusion middleware that bridges precise motion tracking and generative synthesis for general-purpose humanoid control.
MeshMimic bridges 3D scene reconstruction and embodied intelligence to enable humanoid robots to learn coupled motion-terrain interactions directly from monocular video.
EdgeDoG uses DoG kernels and dual uncertainty to improve 3D edge reconstruction.
SOAF models room geometry and wall occlusions for sound propagation.
Puzzles generates novel views and camera paths from a single image or video clips.
DCHM leverages superpixel GS to generate consistent point clouds for label-free multiview detection.
HashPoint accelerates the volume rendering by combining rasterization with ray tracing.
Cardboard human modeling aggregate multiview pedestrian features.
VFA, a voxelized 3D feature aggregation method, improves multiview detection accuracy by reducing occlusion and projection errors.
* denotes equal contribution, † denotes corresponding author.
Multiview detection algorithm design and synthetic dataset generation.
Enabling A Generalized Multimodal Occupancy Perception System on Humanoid.