PhD Researcher at Deakin University

Built physics-based synthetic data pipelines in NVIDIA Isaac Sim and ROS 2, generating millions of labeled manipulation trajectories for grasp-conditioned placement feasibility prediction. Developed geometry-aware learning models for kinematic reachability and collision feasibility, supporting rank-then-plan robotic manipulation under limited planning budgets. Extended to point-cloud-based feasibility prediction without CAD models at test time, using target-conditioned paired-point representations and learned grasp-pose encodings. Expanded the benchmark from a single cuboid setting to multi-category variants across mugs, cans, and cuboids (120 object-scale variants). Designed end-to-end data generation, labeling, training, and simulation-based evaluation workflows in Python, including sim-to-physics transfer analysis. Demonstrated unseen-object grasping with Kinova Gen3 Lite, RGB-D perception, GPD proposals, MoveIt, and hand-eye calibration.

Research Assistant at Monash University

Co-developed a closed-loop robotic manipulation system for in-hand pivoting using vision, wrist force/torque, and tactile sensing. Implemented multimodal control logic for object reorientation while maintaining surface contact and regulating slip. Worked with UR5, Robotiq gripper, RGB-D sensing, and tactile sensing. Contributed to experimental validation showing robust pivoting performance and improved energy efficiency over open-loop and batch-based baselines.

Graduate Teaching Fellow at Deakin University

Teach and support units in data analysis and machine learning through workshops, coding support, and student consultations.