• Architected and deployed robust robotic software for complex HMLV manufacturing, addressing high uncertainty in unstructured environments resulting in high operational uptime.

• Designed and implemented a high-precision calibration methodology for a multi-robot industrial cell, achieving a 10x improvement in accuracy (645 µm mean error).

• Spearheaded the migration of a large-scale industrial robotics platform from ROS 1 to ROS 2 to meet stringent real-time and reliability requirements, implementing lifecycle nodes for robust state management and optimizing critical data pathways.

• Prototyoed a surface reconstruction pipeline for robotic finishing tasks (e.g., sanding, grinding), achieving 1mm average error by fusing multi-sensor data to generate high-fidelity 3D models for accurate toolpath planning.

• Co-developed and published a novel proactive failure prediction system for robotic lift-and-place tasks.

• Engineered a deep learning pipeline that analyzes relative object motion extracted from multi-view RGB streams using SAM2 segmentation and 2D-ICP to predict a continuous risk score, achieving robust generalization to unseen objects in both MuJoCo simulation and real-world experiments.

• Developed a physics-informed learning model to predict screw-tip dynamics, directly mitigating contact uncertainty and enhancing precision for robotic screwdriving in high-mix industrial operations. (Published in RCIM)

• Authored a novel framework for proactive failure prediction in robotic manipulation that generalizes to unseen objects by analyzing relative motion dynamics. (Published in RSS 2025)

• Reduced composite sheet layup cycle time by 50% by engineering a vision-based, adaptive motion planner for 8-DOF robots (KUKA, ABB) that integrated human expertise with data-driven learning. (Published in IEEE CASE 2024)

• Improved robot controller robustness by designing a Quality-Diversity (QD) generative framework that automatically creates diverse 2D/3D environments to systematically stress-test algorithms and expose critical failure modes.

• Developed a hierarchical online planner for human-robot collaboration that proactively adapts to a person's evolving intent by modeling their perceptual limitations (field-of-view) within a custom VR simulation. (Published in IEEE ICRA 2025)