Course Home¶

GT Cloud Robotics is a Vertically Integrated Projects course at Georgia Tech centered around rapid robotics software prototyping. Students pick a focus area and develop a full demo of a working program every two weeks.

This course is about mobile robots in the real world: ground, water, and air vehicles that have to navigate an unpredictable environment and then carry out a long-horizon task. Not self-driving cars, not robot arms. Those live in structured or controlled settings; the hard part here is staying reliable across a long mission in a world that won't hold still.

How the course works¶

1. Figure out what you're aiming at¶

Before you pick a track, research 5-10 companies with open job postings you'd want and could plausibly get. Pull out the skills they're hiring for. Cross-reference Glassdoor and levels.fyi.

The output is a list of skills you can name. Everything after this is you building evidence for them, every two weeks, for a semester.

2. Pick a track¶

At the start, choose one of the four project areas below. This is the space you work in for the semester. You're not committing to a single deliverable; you're committing to a problem area.

Unreal Vehicle Dynamics — Build and test aircraft and boat dynamics in a shared C++ core that runs headlessly or inside Unreal Engine. Work on vehicle models, reference checks, engineering maneuvers, ArduPilot integration, reports, or the boundary between the model and simulator.

Fusion in Motion — Build controlled sensor-fusion experiments for moving ground and aerial vehicles. Work on sensor models, timing and faults, estimator integrations, datasets, metrics, or reports that compare methods against the same measurements and truth.

pipes-rs — Build sensor processing and fusion pipelines in front of ROS 2. Work on sensor inputs, timing and synchronization, camera and point-cloud processing, estimator integration, recording and replay, or testing what happens as sensor rates and data sizes increase.

Gaussian Fieldwork — Explore Gaussian-splat tools on real scenes and work out practical capture and evaluation habits. Compare scenes and reconstruction approaches, document failure modes, and try small robotics uses. This is the most open-ended and research-oriented track.

3. Ship something every two weeks¶

Work runs in two-week cycles. Each cycle you commit to one thing you can demo, and you show it at the Wednesday sync. If you're on a subteam, you commit as a team. Demoing is mandatory. A demo means it works in front of people: a running node, a passing eval, a vehicle doing the thing in sim. Not slides, not a status update.

If it flops, that's fine. You still get up, share your code, and talk through it: what you were going for, where it broke, and why. A demo that fails and gets explained is worth more than a status update that hides the failure. What isn't acceptable is showing up with nothing.

When a cycle closes, that deliverable is done. You pick a new problem to work on, still inside your track. Over a semester you stack up several distinct, shipped pieces of work rather than one open-ended project that's perpetually "almost there."

4. Background modules (optional)¶

The modules below are optional background reading and context to build up a solid foundation. They're here when you want to go deeper on a topic.

Modules:

1. C++ CLI tooling — CMake fundamentals and the workflow to spin up a useful utility from scratch.

2. System design — how robotics services connect, why they run at different rates, and what happens when sensor data stops arriving.

... more coming soon ...