AI-Enabled Tactical Route Planning with Real-Time Human-Robot Collaborations --- Lockheed Martin --- Student Researcher

Technology Used: ROS, PyTorch, Tensor Flow, Mono Depth, Neural Network, Res-Net, Linux, DFS, Fusion Algorithm, AR/VR

• Built a true-to-real-world environment in Unreal Engine (UE4) with drones and cameras, which reduced the cost of doing physical experiments.
• Built communication network and database for fixed cameras and drone cameras which enables real-time distributed target detection.
• Implemented Deep Learning based (Mono Depth) terrain recognition algorithm and Detectron2 algorithm applicable to RGB camera imagery, which are used to generate a depth map for obstacle avoidance and segmentation for classifications respectively.
• Used Rapidly Exploring Random Tree (RRT) algorithm for 2D path planning after achieving the obstacle map from above.
• Designed target tracking strategy for drones and ground robots using robot path planning methods.
• Designed the human-robot collaboration system (LMCO project) for agents tracking as human operators will help drones to update route for unseen obstacles like glass doors or when drones lost the target in foggy and rainy weathers.
• Created collaborative platform between human and robots where robots respond to virtual avatar controlled by human using VR autonomously.
• Analyzed images captured by virtual robot using real time key-point detection algorithm and sent control commands back to the virtual robots which enabled targets tracking and obstacles avoidance.
• Realized interaction of virtual drones and ground robots with physical targets, which is the basis for virtual interaction with the real world.
• Submitted two research papers for review.

Implementing Machine Learning for Enhanced IMCC Heat Condition Simulation --- ASML --- EI Group --- Software Development Engineer

Technology Used: FEA, CAD, Gaussian Acceleration Method, Dense Matrix Computation, Computational Fluid Dynamics, COMSOL, ANSYS

• Employed NeRF for the 3D reconstruction of IMCC and MBI cabinets using multiple images and made fine adjustments using NX.
• Designed an RNN model that transformed case boundary conditions and FEA simulation outputs into precise predictive parameters.
• Integrated FEA, Gaussian Acceleration, and RNN to optimize heat and flow distribution, improving internal layout and elevating heat dissipation efficiency by 18%.
• Certified MBI's flux compliance and found that sidewall wind tunnel performance could increase by 10% with a 1cm thickness enhancement.
• Developed a predictive RNN model for future heat condition assessments, providing proactive heat management for IMCC and MBI cabinets.

Cloud Computations for Distributed Robots Control --- Amazon --- HWI Group --- Software Development Engineer

Technology Used: AWS, Java, Linux, C++

• Implemented and managed cloud computations and communications for robots using Amazon Web Services (AWS), ensuring efficient data exchange.
• Developed and ran path planning (weighted A* algorithm), detection (YOLO), and optimization algorithms on cloud servers to analyze the transmitted images.
• Achieved near real-time communication by returning generated commands from the servers to the robots for execution, thus significantly enhancing their responsiveness.