Why We Invested in Salem Robotics: Autonomy for Hazardous Site Robotics

Why We Invested in Salem Robotics

Autonomy for Hazardous Site Robotics

By Camille Little and Kent Lucas

At nuclear facilities across the United States, there are $200,000 robots collecting dust. They were purchased with the expectation of autonomy. They are not being used, and this story is far more common than the robotics industry would like to admit.

Radiological surveys are federally mandated at every U.S. nuclear facility. Radiation protection technicians walk routes daily, scanning surfaces with handheld instruments, marking paper maps, and manually transcribing data. The process consumes hours of skilled labor every single day. The workforce trained to perform these surveys is retiring faster than it is being replaced, and the robots purchased to address that gap have, almost universally, failed to deliver on their promise.

The problem is not the hardware. Boston Dynamics builds excellent robots. The gap is in the software layer sitting above it. What passes for autonomy today is largely pre-recorded route replay with limited situational awareness, which breaks down quickly in real nuclear facilities with doors, stairs, layout changes, and multi-step workflows. Add regulatory requirements that demand deterministic, auditable outputs, and no existing solution handles the full picture end to end. Facilities find themselves back at a 1:1 robot-to-operator ratio, and the economic case evaporates.

Salem Robotics is building the software layer that closes this gap. Their platform takes robots already on-site and turns them into fully autonomous radiological survey operators, planning routes, executing surveys, navigating obstacles, and auto-generating regulation-compliant records without manual transcription. The core architectural decision is determinism. Where competitors rely on probabilistic AI approaches with variable outputs, Salem builds mathematically guaranteed, auditable motion. In a nuclear environment where an alpha survey requires maintaining a three-quarter inch proximity to surfaces, "probably correct" is not good enough.

What gives Salem a credible path to solving this is the depth of domain knowledge the founders bring. Caleb Horan spent 8-plus years at Los Alamos National Laboratory deploying robotic systems in real nuclear facilities before co-founding the company. Janak Panthi is a PhD researcher whose published work on deterministic manipulation forms the technical foundation of the product. Both hold top-secret clearances. They did not learn this industry in order to start a company. They started a company because they spent years inside the problem and understood precisely what was missing.

Nuclear is the right place to start. With over 200 U.S. facilities operating under mandatory, recurring inspection requirements and substantial budgets to support them, it offers a durable and immediate commercial opportunity. But the longer-term vision is broader. The autonomy stack Salem is building for nuclear translates directly to oil and gas inspection, hazardous chemical environments, biological facilities, and eventually space. The common thread across all of these is straightforward: environments where safety requirements are non-negotiable and the cost of human exposure is too high to ignore.

Several forces are converging to make this the right moment. Nuclear is entering a renaissance, with Morgan Stanley projecting $2.2 trillion in new global investment through 2050. Energy demand is accelerating. The skilled workforce that has historically performed these surveys is shrinking. And across facilities, the hardware investment has already been made. What has been missing is software capable of putting that hardware to work. We are backing Salem because we believe they are the team best positioned to build it.