Bridging the physical and digital worlds—combining my Mechanical Engineering background with Computer Science to build tangible products, intelligent systems, and explore practical AI applications.
Recent Log Entries
Project Logs & Updates
-
2026-09-03
The long way around: I finally finished the computer science degree
I failed a programming course in high school and decided it wasn't for me. Fourteen years and two other degrees later, I finished a BS in CS while working full-time. -
2026-05-01
Introducing GG Swarm Live: From Simulation to the Sky
With the ink finally dry on my Computer Science degree from CSUMB, I’m excited to officially announce the next chapter of my work in autonomous robotics: GG Swarm Live.
During my capstone, I focused heavily on the “brains” of the operation—developing a decentralized coordination framework using Graph Attention Networks (GATv2) and Reinforcement Learning (PPO). While the simulation results in NVIDIA Isaac Lab were promising, the ultimate goal has always been to see these algorithms control physical hardware in the real world.
The Hardware Migration
GG Swarm Live is a dedicated program to transition my research from the “digital twin” to real-world airframes. This isn’t just about porting code; it’s about handling the messy realities of physics, communication latencies, and sensor noise that simulation can only approximate.
The technical stack for this phase is built on industry-standard foundations:
- PX4 Autopilot: The core flight stack for our physical airframes.
- Skybrush: We are integrating our RL-based execution layer with Skybrush to allow for expressive, choreographed drone light shows.
- Crazyflie: Initial sim-to-real baselining will take place on these micro-drones in a controlled indoor environment before scaling to larger outdoor platforms.
Beyond the Capstone
The capstone gave us a v1.0.0 baseline. The roadmap for GG Swarm Live expands this into eight distinct phases, ranging from multi-drone shared-scene training to fully autonomous, obstacle-aware navigation in outdoor environments.
One of the most exciting commercial applications we’re exploring is the “backyard light show” capability. By combining decentralized swarm logic with a user-friendly interface, we hope to make high-end aerial displays accessible for local events and creative displays. Crucially, the decentralized nature of the swarm adds a robust layer of safety; by removing single points of failure, we can ensure more reliable behavior around spectators and audience members.
Follow the Journey
I’ve updated the GG Swarm Project Page to serve as the central hub for this new phase. It now includes the full hardware roadmap and a section archiving the original academic research.
This degree wasn’t a transition to becoming an engineer—I’ve been one since 2012—but rather a significant upgrade to my toolkit. The build is just beginning. Stay tuned for more logs as we take these swarms to the sky.
Gary
-
2026-04-24
Week 16 — Capstone Festival, Program Complete
Last entry. Capstone Festival is today, portfolio and Learning Journal are submitted, and
v1.0.0-capstoneis tagged. CSUMB is done. -
2026-04-21
Week 15 — Project Wrapped, Testing Report Locked, Draft Presentation Cut
Project wrapped for the Capstone Festival. Testing Report closed out, draft presentation video recorded against the Phase 5 cinematic clips, and the README got a final polish including an embedded 8-drone highlight GIF. 3 days remain to the Apr 24 deadline — remaining work is rehearsal, portfolio/journal submission, and the
v1.0.0-capstonetag. -
2026-04-14
Week 14 — Phase 5 Complete, Phase 6 in Flight, Thesis Clarified
Phase 5 done 12 days ahead of the M4 gate, Phase 6 underway. The cinematic trailer was edited from the 20 captured clips and published to YouTube early in the week. The back half shifted to documentation and framing — making the research question explicit: coordination, not stabilization. 10 days remain to the Apr 24 deadline. No training is planned; only editing, rehearsal, and submission remain.
-
2026-04-07
Week 13 — Phase 4 Complete, Showcase Prep Begins 7 Days Early
Forest navigation now works cleanly — 0 body penetrations in 700-step play with 8 drones weaving through 40cm tree trunks. Phase 4 is complete and Phase 5 (Showcase Prep) starts 7 days early.
-
2026-03-31
Week 12 — Phase 3 Complete, 9 Days Early
Rebuilt Phase 2 from scratch, achieved formation control in a single session, then blew through Phase 3 — finishing 9 days ahead of schedule.
-
2026-03-24
Week 11 — Phase 2 Complete, Formation Control Achieved
Completed Phase 2 — drones can now maintain stability and hold formation distance from each other.
-
2026-03-21
Infrastructure Upgrade: Migrating from DigitalOcean to Google Cloud Run
After running on a DigitalOcean droplet for some time, I made the decision to migrate this portfolio site to Google Cloud Run. This wasn’t just a technical exercise—it was driven by practical considerations around cost, scalability, and operational simplicity.
-
2026-03-17
Week 10 — MAPPO Training Pipeline and Environment Finalization
Finalized the MARL environment, set up the MAPPO training pipeline, and replanned the schedule after falling ~2 weeks behind.