Dr. Sertac Karaman has spent two decades teaching machines to move faster and think smarter than the humans who built them. Now, as founder of Tycho.AI and director of MIT’s Laboratory for Information and Decision Systems, he has moved that research out of the lab to fly in some of the most contested airspace on earth. On a recent episode of the Dawn of Autonomy podcast, Karaman explained why his company believes small, fast, GPS-independent drones represent the next real chapter in unmanned systems for defense, and why the timing has never been better.
The Inspiration of DARPA’s Urban Challenge
Karaman’s path to Tycho.AI began long before drones dominated headlines. He holds a master’s in mechanical engineering and a PhD in computer science from MIT. In 2007, he competed in the DARPA Urban Challenge, a contest that pushed self-driving cars from science fiction toward reality. That inspired him to found Optimus Ride, an autonomous shuttle company that grew to roughly 250 employees before Magna Electronics acquired it.
Looking back at those early years, Karaman marveled at how much confidence the field had despite how little actually worked. “We were so optimistic that this thing would work, but nothing worked back in the day,” he said. That optimism, he explained, came from a belief that autonomy could eventually outperform any human operator, a concept he referred to as “super autonomous vehicles.”
That belief fueled an obsession with high-performance autonomy. He led some of MIT’s early drone racing experiments, where his team became one of the first groups to fly autonomous drones faster than human FPV pilots through obstacle courses, complete with backflips through gates. At the time, he had to explain basic FPV (first person view) concepts to defense leaders who had never seen such a drone in action. Then everything changed.
2022 Changed Everything
For years, Karaman struggled to recruit MIT talent into defense work. Engineers wanted to build self-driving cars or consumer robotics, not weapons systems. That resistance evaporated almost overnight when Russia invaded Ukraine. He watched as Ukraine’s engineers and citizens improvised battlefield technology under existential pressure. MIT researchers who had once dismissed defense work suddenly volunteered to work with him. “2022 changed everything,” he said, reflecting on the sea change in sentiment.
The instinct to support an outmatched defender in a righteous cause accounted for only part of the equation. Karaman added the sheer difficulty of the problem sets also intrigued MIT engineers. Building a drone that flies fully autonomously for counter-UAS missions, for example, demands solving perception, navigation and control problems simultaneously, without any margin for error.
The following year, Tycho.AI licensed intellectual property out of MIT and began building what would become its Voyager autonomy stack. Even then, investors were skeptical of defense-focused startups. Karaman recalled pitching venture capitalists who openly questioned why anyone would build a company that sells to the government. Now, just a few years later, defense autonomy has become one of the most heavily funded categories in venture capital, and Tycho.AI sits at its technical frontier.
Defining Agile Autonomy
When asked what Tycho.AI actually builds, Karaman offered a single phrase: “agile autonomy.” He described it as an extension of the super autonomous vehicle concept, engineered specifically for vehicles that must detect and avoid obstacles at high speed, target other aircraft in flight and maneuver with a responsiveness no human pilot could match.
Karaman believes that most drone development still follows an outdated blueprint: take a piloted aircraft, remove the pilot, then shrink the airframe. Tycho.AI inverts that logic. “We build intelligence in a box that’s as small as possible, and then we build a vehicle around it,” he said. That intelligence, he added, is not shackled by the biological limits that constrain human pilots, which frees the airframe to behave in ways a manned aircraft never could.
Tycho.AI targets the specific gap in the market for low-altitude, high-speed autonomy. It designs for the kind of maneuvering associated with army aviation, rather than the thousand-foot cruising altitude most commercial drones occupy today. Karaman believes that gap will only widen in relevance as adversaries develop better detection systems against aircraft flying at predictable altitudes.
3 Pillars Behind Tycho.AI
Tycho.AI’s central differentiator sits at the convergence of software, silicon, and hardware as the company organizes its technical advantage around three core competencies:
- AI and autonomy software: Karaman positions his team as a defense-focused AI and autonomy lab applying cutting-edge computer vision research directly to flight control.
- Chip design: Tycho.AI engineers have developed custom chip sets to maximize data throughput to onboard GPUs, with plans to develop dedicated co-processors over time.
- Air vehicle design: The company has already built an airframe capable of precision agile maneuvers at speeds reaching 200 miles per hour, manufactured through 3D printing and injection molding rather than costly carbon fiber layups.
“The more we integrate these three core competencies, the more high-performance systems we’re going to develop,” Karaman said. The platform born from this philosophy is Halley, a Group 1 VTOL aircraft that officially debuted at SOF Week 2026.
Meet Halley: A New Category of Small UAS
Viewed head-on, the Halley airframe resembles the letter H, built around four propellers paired with both horizontal and vertical wings. Karaman explained that the vertical wings solve a problem plaguing most quadcopter-style designs: the ability to change orientation instantly without sliding sideways through the air like a boat. “The moment you turn it, it’s looking at another place,” he said. That capability proves essential for intercepting moving targets or dodging obstacles at speed.
At just 5.2 pounds, with foldable wings and zero-tool assembly, Halley carries a custom autopilot built on nonlinear control systems developed through Karaman’s MIT research, rather than relying on open-source flight controllers like ArduPilot. Karaman was quick to note that raw speed without endurance means little on an actual mission. Competing platforms boast headline speeds but negligible operational range. “If it goes for two minutes, what do I do with that?” he said. Halley, by contrast, sustains 20 minutes of flight time while executing the aggressive maneuvers for which Tycho.AI built it.
Proving the Concept at T-REX 26-2
Tycho.AI validated Halley’s capabilities at Camp Atterbury’s T-REX 26-2 event, an independent Department of Defense evaluation held in May. There, the team demonstrated high-altitude climbs, live intercepts against foam aerial targets and positional control without GPS, a critical capability for denied environments where satellite navigation is jammed or unavailable. Karaman also revealed, without fully unveiling it, that Tycho.AI transmitted live camera imagery over extremely limited bandwidth during the event. That feat reportedly left evaluators trying to figure out how the data made it through at all.
Karaman emphasized why GPS-denied positional control matters so much for scaling these systems to a broader operator base. Expert drone racing pilots can compensate for a lack of automated stabilization, but the military increasingly needs 20-something operators with minimal training to safely fly aircraft carrying live payloads at 200 miles per hour. Voyager’s autonomy stack, he explained, closes that gap by handling the flight dynamics that would otherwise require years of piloting experience.
Security also matters as much as performance in this equation. Tycho.AI recently achieved CMMC Level 2 certification, a rigorous third-party audit of cybersecurity practices required for deeper Department of War engagement. “I know the weight of the stuff that we’re building, and I think I want to make sure that we’re careful and safe with it, as we build this for our country,” Karaman said. He views certification as a moral obligation rather than simply a bureaucratic checkbox.
Scaling Agile Autonomy Across the Fleet
The Tycho.AI team designed its Voyager autonomy stack to extend beyond Halley. Karaman described a hardware and software stack capable of assuming full control of an aircraft’s control surfaces or propellers, executing genuine formation flight with aircraft just ten feet apart, and deploying across Group 1 through Group 5 platforms. He distinguishes this from typical swarm coordination, which often amounts to simple task assignment. Tycho.AI’s approach demands tight, real-time coordinated maneuvering instead.
For a company that Karaman jokingly described as “flying under the radar,” quite literally given Halley’s low-altitude design philosophy, Tycho.AI’s next moves promise to make plenty of noise. With partnerships still in the pipeline and government With partnerships still in the pipeline and government demonstrations already underway, Karaman was candid about commercialization timelines. When asked when that will begin in earnest, he said, “Now.” He noted that the company has several partnerships still awaiting public announcement.
That sense of urgency reflects a broader shift in how Karaman thinks about scale. Manufacturing thousands of units, not just prototypes for evaluation, now presents the central challenge. He envisions a supply chain that runs from raw silicon to finished aircraft entirely within U.S. borders, paired with production processes simple enough that the same factory floor building consumer electronics could just as easily turn out 200-mile-per-hour intercept drones.
Karaman also made a direct appeal to talent. Tycho.AI, based on MIT’s campus, is actively recruiting engineers who want to work on problems that matter. “We’re excited to attract talent into our company,” he said, noting the team includes drone racing champions and researchers who have spent careers pushing the boundaries of autonomous flight.
For Karaman, the mission comes down to the singular belief that agile autonomy isn’t a feature bolted onto existing drones. It’s an entirely new category of unmanned systems. And Tycho.AI intends to define it.
Watch Dr. Sertac Karaman on Episode 132 of the Dawn of Autonomy.