United States
I like computer algorithms and math that make robots do cool things
Lead an engineering team, UCI UAV Forge, of 50 members, to placement at SUAS International Competition. Oversaw Guidance Navigation & Control, Computer Vision, and Avionics teams. Managed systems design and integration.
Developed and deployed real-time computer vision pipelines for autonomous UAV missions under strict embedded hardware constraints. Designed an imaging and inference pipeline optimized for onboard processors using GPU-accelerated, memory-constrained compilation. Integrated Google’s EfficientDet for shape-cutout detection during autonomous flight. Implemented and optimized K-means–based color and letter classification for aerial target recognition. Delivered real-time inference performance through runtime and memory optimizations on embedded GPUs. Directed a team of engineers to combine machine learning and classical vision algorithms for competition-grade perception systems. Purpose: enable autonomous aerial target detection and identification for competition objectives.
As a research speaker at the American Astronautical Society (AAS) Guidance, Navigation, and Control (GNC) Conference, I had the privilege to present my results and work in the field of autonomous spacecraft rendezvous and docking. My presentation focused on "Learning to Plan ARPOD (Autonomous Rendezvous and Proximity Operations Demonstration) via model-free Reinforcement Learning" where I demonstrated successful results in significantly more challenging conditions compared to similar attempts at applying RL in this field. Additionally when compared to conventional control, my results demonstrated an ability to simplify the procedure using neural networks by merging independent mission stages into a unified control law that produces overall smoother trajectories. During the conference, I shared my research findings and insights on leveraging Reinforcement Learning (RL) algorithms to optimize the trajectory planning and control of autonomous spacecraft during rendezvous and docking maneuvers. This groundbreaking work, approved by Springer Nature for publication, highlights the application of RL in solving complex decision-making problems in the context of space exploration. Through the utilization of RL techniques, I trained a model to perform spacecraft docking using Proximal Policy Optimization (PPO) algorithms. I developed a dynamic simulation environment that accurately represents the 3-body orbital dynamics involved in the docking process. Additionally, I optimized the model-free PPO RL control module to enhance the efficiency and effectiveness of the docking procedure.
Research work in collaboration with the Air Force Research Laboratory under UCI Under Graduate Research Program.