Graduate Research Assistant
Internal Proprietary ToolingImageBounding Box
Developed hybrid ML methods for robotic manipulation and motion planning, improving stacking and grasping performance toward human-like dexterity.
Built a 3D block pose estimation pipeline by combining SAM-based segmentation with CREStereo stereo depth, reaching ~99% pose accuracy for stacking and grasping tasks.
Integrated the pose estimator into a ROS2-controlled robotic arm and used Isaac Sim (including diffusion-based scene variations) to generate synthetic stacking scenarios and test perception and motion strategies before deployment on the real robot.
Developed hybrid ML methods for robotic manipulation and motion planning, improving stacking and grasping performance toward human-like dexterity. Built a 3D block pose estimation pipeline by combining SAM-based segmentation with CREStereo stereo depth, reaching ~99% pose accuracy for stacking and grasping tasks. Integrated the pose estimator into a ROS2-controlled robotic arm and used Isaac Sim (including diffusion-based scene variations) to generate synthetic stacking scenarios and test perception and motion strategies before deployment on the real robot.
2023 - 2025