Projects

Implemented and compared Extended Kalman Filter (EKF) and Particle Filter algorithms for robot localization on the PR2 robot. Tested under five different challenging scenarios with complex paths to study advantages and limitations of each approach. Analyzed performance in terms of accuracy, computational efficiency, and robustness to sensor noise.

Developed a complete autonomous navigation system for a differential-drive mobile robot using ROS2 and the Nav2 navigation stack. Implemented real-time obstacle detection and collision avoidance in C++ by processing lidar data, enabling safe navigation in dynamic environments. Integrated SLAM for mapping and localization.

Conducted a comprehensive analysis of failure cases in state-of-the-art multi-view 3D reconstruction method MASt3R under challenging capture conditions. Collected and annotated a custom scene dataset with varying lighting, occlusions, and viewpoint distributions. Documented systematic limitations and proposed potential improvements for future work.

Built a classical computer vision pipeline to detect barcodes on grocery items and classify bounding boxes by barcode readability status. Used traditional CV techniques including edge detection, morphological operations, and contour analysis. Achieved robust detection across various lighting conditions and barcode orientations.

Created a benchmark audio dataset with 1M+ samples using multiple TTS models with paired real/fake audios. Engineered end-to-end pipelines for generation, labeling, and evaluation. Trained an audio-language detection model that outperforms human listeners and achieves state-of-the-art performance on deepfake detection.

Addressed activity hallucinations in Video-LLMs by creating an annotated video-activity dataset with fine-grained evaluation metrics for actor, action, object, and environment understanding. Designed targeted question types and evaluation methods for robust activity recognition.