Opening the Next Robotics Chapter

Building for the Battlefield

A few years ago, halfway through my PhD, my advisor, Ufuk Topcu, and I decided to build something. No external funding, no institutional backing — just a conviction that multi-agent artificial intelligence systems paired with formal mathematical methods could meaningfully change how the military plans and executes operations, specifically for autonomous robotics systems. We started with autonomous mission planning for unmanned aerial and ground vehicles, a problem centered on getting algorithms to reason over complex mission constraints, asset capabilities, and dynamic environments without a human micromanaging every decision (see Autonomous Mission Planning for Unmanned Robotics Systems).

The military’s appetite, however, pointed us in a different direction. The early stages of the war in Ukraine were just beginning to reveal how consequential the large-scale deployment of autonomous systems — drones in particular — would become on the modern frontline, but that understanding was not yet widespread in the military, or at least not with our military connections. What our connetions were mostly interested in was large-scale fleet coordination, specifically the ability to reason over and manage contested logistics scenarios, where supply chains, routes, and resources operate under adversarial pressure. We followed that signal. Our tools found a particularly enthusiastic user base among wargamers, professionals who stress-test military plans in simulated environments. Watching operators use our software to untangle genuinely complex logistics puzzles and develop a sharper understanding of their scenarios was among the most satisfying moments of the company (see Multi-Agent Decision-Support in Adversarial Environments).

The Grind of Building a Defense Startup

We bootstrapped from zero to seven-figure funding. For those unfamiliar with defense contracting, that means navigating Small Business Innovation Research (SBIR) grants, direct contracts, and a procurement ecosystem with its own rhythms and vocabulary — none of which I knew when we started. I came straight from university, with no industry experience and no roadmap. As an international student on an F-1 visa from Germany, I was doing all of this while navigating the immigration constraints that come with building a company in the United States — a layer of complexity that made an already steep learning curve feel, at times, damn near vertical. The operational load — proposals, contracts, compliance, hiring — sits entirely outside the technical work, and for long stretches it consumed most of my hours. I’m grateful to the team at SBIRAdvisors, particularly Clint, whose expertise in navigating the defense contracting landscape saved us significant time and effort on multiple occasions.

None of it would have happened without the people I got to work with. The late-night sprints before Broad Agency Announcement (BAA) submission deadlines, the debugging sessions that stretched past midnight, the particular camaraderie that forms when a small team is pushing hard toward a shared deadline — I’m deeply grateful to my co-founders Matt and Ufuk and the rest of the team for bringing that energy, consistently.

We ran a successful business that made it further than most defense startups do. We had deep connections in the Pentagon and got to seven-figure funding within just 2 years. But the path had its fair share of structural obstacles that made scaling genuinely difficult: bureaucracy, ever-shifting military directives and priorities, security clearance requirements, and — most consequentially — painfully long contract cycles that demanded a constant re-evaluation of whether the time invested would yield meaningful returns in the mid- to long-term. Beyond the structural friction, the longer I ran the company, the more I felt the pull back toward robotics, specifically the hands-on, systems-level challenge of building machines that perceive and act in the physical world. We were operating in an AI space adjacent to robotics, but had drifted considerably from the autonomous mission planning of robotic systems I had originally set out to pursue.

After almost four years of navigating those scaling challenges, and with my passion for robotics not subsiding, the team arrived at a natural inflection point: was this the path we wanted to continue on? We concluded that we each wanted to start new chapters. For me, that meant one thing; returning to the robotics frontier.

Back to the Robotics Frontier

Throughout the company, I stayed technical by contributing directly to the engineering side. But I knew that returning to the frontier of robotics research required more than staying technical at the margins — it required a return to a research-first environment, where hard problems can be pursued with depth and rigor.

I have recently started a postdoc at the University of Texas at Austin, within the Center for Autonomy and the Texas Robotics program. My focus going forward is on challenging, general-purpose robotics problems — particularly robot learning and its deployment in real-world environments, where the gap between what works in simulation and what holds up under physical reality remains one of the field’s most demanding open questions. I’m grateful to be contributing to the robotics frontier within one of the strongest robotics ecosystems in the world, and look forward to continuing that journey in industry again soon enough.