At Tech, drones are more than just a weekend hobby; they’re a portal to a science fiction future that might not be as far away as most people expect. Across campus, labs and Vertically-Integrated Project (VIP) teams, Tech faculty are driving cutting-edge endeavors. They are solving the greatest challenges that come with drone technology: how autonomous systems see a changing world, predicting what comes next and how to safely move through it.
“We don’t predict one future, we plan for many,” said Panagiotis Tsiotras, associate director at the Institute for Robotics and Intelligent Machines.
Tsiotras has spent more than 25 years at Tech developing algorithms for path planning and trajectory optimization. He tries to find the fastest and most efficient path from A to B. His team at the Autonomous Vehicle Safety VIP creates four-dimensional pictures of spaces, using the standard three dimensions and adding time. This “multi-future” planning allows them to predict how the world will evolve a few seconds into the future and then choose a safe and efficient path.
The algorithms propagate many plausible futures for cars, cyclists or pedestrians and select a trajectory that avoids collisions. In practice, this means integrating perception with control and optimization, allowing the vehicle to react smoothly when unexpected events occur.
Tsiotras explained that AI today is used “primarily on vision recognition tasks and creating the 3D — really 4D — representation of the environment,” and planners then “predict not only one future, but different futures … and take the best step forward.”
This theory drives the software of Tech’s autonomous vehicle research. The practice lives a few buildings away, where Experimental Flights lead, Michael Mayo asks a practical question: What useful job can a drone do this week, safely and reliably?
Mayo has taught Experimental Flights since 2017, steering the VIP into drones around 2019. His credibility comes from turning autonomy into end-to-end, working missions. The class’s early flagship project was a campus delivery prototype, a quadcopter tied to a smart-locker drop-off and a live-tracking app. It proved the full stack (hardware, autonomy and user experience), even if running a real network would require FAA waivers and dedicated operators.
Today, the team runs parallel mission tracks that map cleanly onto real needs: wildfire support, where small unmanned aerial vehicles (UAV) use sensors for reconnaissance and basic fire-spread prediction (intelligence, not water drops); ad-hoc communications relays, with drones acting as radio repeaters in mountainous terrain or disaster zones to restore links before ground crews arrive; indoor inventory, using collision-aware UAVs to scan QR codes along warehouse aisles and update stock — no ladders, no clipboards; and disaster relief, deploying aircraft to find people, maintain comms and deliver light supplies when roads are washed out.
Where Tsiotras stresses guarantees and theory, Mayo focuses on engineered operations and constraints. Students design around community acceptance — they do their utmost to shrink the noise footprint, protect privacy by design and operate within Federal Aviation Administration (FAA) rules and documented risk mitigations. They publish flight windows and routes, geofence sensitive areas, keep hover time near people to a minimum, and build redundancy and remote-pilot takeover into missions so they can point to a concrete safety case. The goal isn’t just to fly — it’s to fly in a way neighbors can live with, so the program earns the trust needed for tougher missions and longer corridors. Mayo is also frank about the biggest bottleneck for delivery-style concepts.
“If you need someone to have eyes on the drone, that person could have walked the package,” said Mayo.
He believes that the pathway to scale lies in building systems with redundancy, setting up remote-pilot takeover stations and assembling strong safety cases to earn targeted waivers in controlled corridors. That might look like having drones fly up to windows late at night to deliver forgotten parts or missing tools that one would not wish to walk all the way to grab.
Just like Tsiotras’s group, Mayo’s team uses AI, where it helps most right now — perception and simple prediction to make mission-level decisions.
Finally, both instructors point students to clear on-ramps. They encouraged students interested in drone technology concepts to pursue AE, ECE, CS and MATH departments like control, optimization, discrete algorithms, stats/ML, and prove it with data and edge cases.
Those who seek hands-on experiments can join clubs like Experimental Flights to build, fly and iterate, or they can look to Georgia Tech Research Institute internships/co-ops where modeling and simulation, flight test and multi-vehicle coordination are applied to research projects. The two tracks are different steps towards entering that portal into a science fiction reality where autonomous systems work tirelessly at the beck and call of everyday people and where one person can do the work of 10.