Research Experience

Developed real-time vision and pose-estimation pipelines for autonomous perching, with hardware-in-the-loop validation of approach and contact behavior.

• Developed real-time detection and pose-estimation pipelines from camera/sensor specifications, delivering target state estimates for onboard perching control
• Designed and validated approach-trajectory and contact-dynamics controllers for repeatable perching maneuvers
• Contributed to "Perch: A Vision-Based Approach for Autonomous Perching," under review for IEEE/RSJ IROS 2026

Developing safety-critical control systems for humanoid robots using RGB-D perception and real-time system identification.

• Engineered an end-to-end RGB-D perception pipeline (capture → point cloud filtering → feature extraction) delivering robust shape/pose priors to accelerate controller inertia estimation for humanoid SATYRR
• Implemented center snapping and geometry estimation by segmenting point clouds from depth data; produced consistent results across varying lighting/occlusion conditions
• Customized NVIDIA Jetson Orin Nano environment (device tree overlays, driver configuration, udev/systemd setup) enabling real-time sensing and motor interfacing on an embedded humanoid platform

Developing traversability prediction systems for autonomous navigation in semi-structured environments using weakly-supervised learning approaches.

• Tuned MPPI controller (ROS Noetic / Gazebo) and packaged improvements into a reusable ROS module for traversability-aware planning
• Collected multi-environment datasets (campus + field) and prepared weakly-labeled training corpus; trained a neural model to refine sampling for safer navigation
• Investigated backtracking strategies for low-supervision systems in semi-structured agricultural terrains
• Self-studied and integrated ROS, PyTorch, and Gazebo toolchains for rapid experimentation