Leading a NASA Robotics Team to World Championship

In 2024, I had the incredible opportunity to lead a 5-person software team for the NASA Robotics Alliance Project with Pasadena City College’s Lancer Lumineers. We competed against 79 teams worldwide and secured 13th place at the world championship.

The Challenge

Our mission was to build and program a Remotely Operated Vehicle (ROV) capable of performing complex underwater tasks. The key challenges were:

• Real-time human-robot interaction
• Underwater navigation and manipulation
• Telemetry streaming and command processing
• 3D reconstruction of targets

What We Built

I optimized the GUI for pilot feedback with real-time commands and telemetry streaming, which improved ROV maneuverability by 30% during operations. We also integrated 3D reconstruction capabilities to generate accurate real-world scaled models for engineering validation.

Current Work: Network Systems for Culvert Inspection

Today, I’m working as a Robotics Engineer Intern at Caltrans, developing an end-to-end control stack for a culvert-inspection robot. The system operates over custom-made IP Manet radio and exposes a RESTful API built with FastAPI.

Technical Highlights

• WebSocket endpoints running at 20-50 Hz for real-time control
• Field-ready network topology enabling simultaneous HoverMap and RaspberryPi access
• Automated network bring-up with Bash/Python scripts for health checks
• DHCP/static addressing, routing rules, and NAT/port-forwarding configuration

The automation scripts I developed cut on-site setup time significantly and accelerate fault isolation using ping, iperf3, and tcpdump logging.

Machine Learning for Autonomous Vehicles

At UC San Diego’s SEElab, I’m pushing the boundaries of trajectory prediction for autonomous vehicles. I’m developing and benchmarking novel models including:

• A dual-head Graph Neural Network (GNN)
• An LLM-based approach using LLaMA-Factory

Data Engineering at Scale

I architected a Python ETL pipeline using pandas and NumPy to process 100k+ multimodal NuScenes records including LiDAR, camera, and CAN bus data. This pipeline enables context-aware decision making for autonomous vehicles.

I’m also designing an agent-centric framework for 3D dense captioning in AVs, defining system requirements for dynamic scene understanding.

---

These experiences have taught me that robotics isn’t just about writing code—it’s about building reliable systems that work in the real world, under real constraints.