Juncheng Chen is a graduate student in the FAST-FIRE Laboratory. His research interests lie in 3D vision across multiple views and sensors. During the first stage of his master’s studies, he focused on depth completion for ToF cameras in large-scale scenes (sensor-based 3D vision). In the second stage, his work shifted toward feed-forward 3D Gaussian Splatting (3DGS) tasks, including 3DGS-aware image matching and real-time feed-forward 3DGS reconstruction (multi-view 3D vision). His algorithm design emphasizes deployability in downstream robotic tasks.

So far, he has completed two projects exploring a unified Real–Sim–Real methodology: data are first collected in the real world and reconstructed into high-fidelity realistic representations to generate training data, which are then used to train models applicable to real-world environments. The Real-to-Sim process brings the simulator closer to reality, providing greater flexibility and richer supervision signals for Sim-to-Real training within the simulator.