Investigating memory architectures that preserve salient spatiotemporal context for coherent long-horizon video generation. Designing memory update and retrieval strategies to reduce temporal drift and maintain consistent identities, motion, and scene structure across extended sequences. Supervised by Yu-Cheng Chou and Prof. Alan Yuille.

Developed a geometry-guided pipeline that lifts exocentric frames into 3D and warps them along target head-camera trajectories to provide partially observed egocentric priors. Fine-tuned a pretrained video-diffusion model with LoRA and autoregressive history conditioning, enabling coherent long-horizon egocentric video generation. Supervised by Liyang Wu and Prof. Alexandre Alahi.

Implemented a bridge-worker framework on Stable Diffusion 1.5, supporting latent-window slicing, spherical prompt interpolation, and multi-prompt panoramic composition. Evaluated Bridge Correction and SMC/FKC-PoE sampling across 50+ configurations, improving the reliability of seamless transitions between generated regions. (EPFL CS-503 Visual Intelligence)

Developed a real-time state synchronization and visualization system in Unity using the Default Visualization Suite, integrating scans, maps, costmaps, route plans, and goal tracking. Implemented a hybrid obstacle-avoidance framework combining ROS2 SLAM/Nav stack and Unity NavMesh with runtime switching and goal-skip logic for unreachable targets.