At nuclear facilities across the United States, a particular species of expensive hardware has taken up residence in storage closets and equipment bays. Boston Dynamics Spots. ANYmal quadrupeds. Various wheeled platforms that were supposed to revolutionize radiation safety. The machines boot up just fine. They can walk, roll, navigate. What they can't do—at least not without a human tethered to a control station—is the actual job they were bought for.
The problem isn't the robots themselves. It's that someone forgot to build the brain.
Salem Robotics, a two-person startup emerging from Y Combinator's Summer 2026 batch, is betting it can fill that gap with software alone. The Austin-based company launched publicly in fall 2026 with a proposition that sounds almost too straightforward: take the robots nuclear operators already own, add Salem's autonomy layer, and suddenly those expensive machines can run radiation surveys on their own, generating compliance documentation that satisfies regulators and eliminates human exposure hours.
Whether it works remains to be proven. But the pitch addresses something real—a peculiar mismatch between capital expenditure and actual utility that's become a quiet embarrassment in the nuclear industry.
Autonomy, Not Hardware
Salem's founders, Caleb Horan and Janak Panthi, aren't selling robots. They're selling the software to make existing robots useful.
Their platform connects with radiation-capable instruments already standard in the industry, autonomously plans inspection routes, executes the survey missions, and outputs time-stamped, spatially linked data formatted for regulatory review. The emphasis throughout is on deterministic manipulation—precise, repeatable movements with what the founders describe as "theoretical safety guarantees"—rather than the probabilistic AI approaches that can falter in safety-critical environments.
"Contact-rich tasks," in the company's framing: opening doors, turning valves, maneuvering sensors into tight spaces. The kind of physical interactions that sound mundane but become complicated when errors might mean radiation exposure or a failed inspection that triggers regulatory scrutiny.
The bigger ambition, though perhaps the more difficult sell, is replacing the paper-based workflows that still define routine radiological surveys. Radiation protection technicians at commercial plants walk predetermined routes with handheld detectors, clipboards, and maps. They transcribe readings. They file reports. It's methodical, compliance-driven work that hasn't fundamentally changed in decades, even as facilities have spent six figures on mobile robots.
Salem's dashboard lets operators plan missions remotely, monitor progress, and receive structured documentation afterward—deliverables designed to satisfy the Nuclear Regulatory Commission's exacting recordkeeping requirements. If it works as advertised, the software becomes the missing link between expensive hardware and actual labor savings.
Academic Roots in a Commercial Package
Horan and Panthi met at the University of Texas at Austin's Nuclear and Applied Robotics Group, one of the few academic labs in the country focused specifically on robots for radiological environments. Horan, now CEO, spent nearly a decade in the nuclear industry, including five years as a year-round intern at Los Alamos National Laboratory working on autonomous radiation inspection systems. Panthi, the CTO, directed his PhD research toward deterministic manipulation and quadruped robots performing contact-intensive survey tasks in nuclear settings.
The academic pedigree isn't incidental. A preprint the pair co-authored in early 2024 with their advisor, Mitchell Pryor, explored collaborative mobile robots conducting radiation surveys in active facilities—research that appears to have translated fairly directly into Salem's commercial offering. Pryor, who brings more than 20 years of robotics experience in nuclear and hazardous environments, remains an advisor.
The company secured a $25,000 investment prize from non sibi ventures at the Rice Business Plan Competition earlier this year, then entered Y Combinator. YC's standard deal provides $500,000 for participating companies; non sibi has also invested, though the company hasn't disclosed total funding raised.
A Market That Bought Hardware Without a Plan
Salem's timing reflects—or perhaps exploits—a broader pattern in how nuclear operators approach automation. Boston Dynamics has been public about robot deployments at sites like Dominion Energy's facilities, the UK's Sellafield complex, and Fukushima Daiichi. ANYbotics has placed quadrupeds in Finland's Onkalo deep geological repository and partnered with UK integrator Createc for facility inspections.
Most of those deployments, however, center on teleoperated data collection: a human pilot drives the robot around, the robot captures sensor data, and someone still processes everything manually on the back end. Some systems handle narrow autonomous tasks—following a predetermined path in a low-risk environment, for instance—but fall short of the full inspection workflow that would genuinely replace a technician's survey route.
That gap is what non sibi ventures highlighted in an investment thesis earlier this year. The firm estimated more than 200 U.S. nuclear facilities face recurring mandatory inspection requirements, representing what it characterized as a substantial automation opportunity largely unrealized due to software limitations rather than hardware constraints.
Salem also lists oil and gas operations, CBRN (chemical, biological, radiological, nuclear) response, and space infrastructure as potential expansion markets. For now, though, the focus remains nuclear—a sector with high barriers to entry but equally high switching costs once a vendor proves it can deliver.
Validation Over Revenue
The company remains in what it's calling validation mode. Salem is offering short-term engagements—one to two weeks, scoped narrowly—to prospective customers, but it hasn't named any clients or disclosed revenue. The team is still just Horan and Panthi, according to Y Combinator's public directory, along with a small set of advisors including Pryor and Jim Nelson, co-founder of the UK insurtech Simply Business.
That lean structure may be intentional for a pre-revenue startup, or it may reflect the challenge of scaling in a market where sales cycles are long and risk tolerance is low. Nuclear operators don't move quickly. They don't experiment casually. Proving that software can generate compliance-grade outputs—outputs that will withstand NRC review and genuinely reduce radiation dose to personnel—requires time, testing, and a level of regulatory scrutiny that venture-backed startups don't typically encounter.
The product's positioning as hardware-agnostic is both a strength and a vulnerability. Salem doesn't have to manufacture or support robots, which keeps capital requirements low. But it also means the company depends on facilities already owning compatible platforms—and on those platforms being reliable enough to stake safety-critical operations on.
Whether nuclear operators will trust a two-person startup with compliance workflows that carry regulatory and safety implications is the unanswered question. Salem's UT Austin research background and deterministic technical approach give it credibility in a field where academic rigor still matters. Now comes the harder part: proving the commercial model, one cautious validation engagement at a time.
The $200,000 paperweights, after all, were supposed to be the easy part.
