Visual Servoing with Residual Reinforcement Learning
Combining classical IBVS with a PPO residual policy for robust eye-in-hand manipulation on a Franka Panda robot.
Projects
Audio-Driven Lip Synchronization
A production-grade pipeline for generating photorealistic lip-synced face videos from arbitrary speech audio, combining SyncNet with a U-Net generator.
Medical Paper Abstract Structurization
Fine-tuning LLaMA with LoRA to convert free-text medical abstracts into structured JSON — covering synthetic data generation, multi-layer quality filtering, targeted loss design, and field-level F1 evaluation.
3D Motion Planning
Comparison of Bi-directional RRT and Weighted A* for collision-free path planning across seven 3D obstacle environments.
Visual-Inertial SLAM
An EKF-based visual-inertial SLAM pipeline that fuses high-rate IMU prediction with stereo observations for 3D trajectory and landmark mapping.
VLA Data Efficiency Study on LIBERO
Fine-tuning SmolVLA across 5 data regimes on LIBERO-Spatial to investigate how many robot demonstrations a Vision-Language-Action model actually needs.