FieldAI builds robot foundation models for machines that work outdoors and underground, where GPS drops out and no prior map exists. The company is based in Mission Viejo, California and was founded by Ali Agha, Shayegan Omidshafiei and David Fan out of research work that runs back to 2016. Its Field Foundation Models and EDGE general-purpose robot brain are deliberately robot-agnostic, running on legged, wheeled, tracked and flying platforms rather than being written for one machine geometry, with risk-aware navigation handled on the edge. Deployments span three continents. In September 2026 Caterpillar announced a collaboration to bring FieldAI’s models to Cat machines for autonomous inspections, jobsite and facility digital twins, and earlier risk detection. Other construction partners include McLaren Construction Group and Ryan Companies.
The technical claim underneath FieldAI’s product is narrower and more interesting than general-purpose autonomy. Its Field Foundation Models are built around a Belief World Model, described as the predictive engine that reasons under uncertainty and lets a robot keep operating reliably when something unexpected happens in a real environment. The company calls this class of system risk-aware robotics AI, and the distinction matters: the design assumption is not that the world is known, but that the robot’s belief about it is incomplete.
That premise drives the second design choice. Rather than writing autonomy for a single machine geometry, FieldAI positions one brain for any machine, with the same models running across different robot types, tasks and environments. The EDGE product is the packaging of that idea.
The team behind it is drawn from DeepMind, NASA JPL, Tesla, NVIDIA and Amazon, and the company describes deployments as field-proven across three continents with unlimited use cases, working with global industry leaders and further strategic partnerships still unannounced.
Four markets are named on the company’s own site: construction, urban operations, industrial and energy, and federal. Construction is treated as a first-class vertical rather than a demonstration case, and the published work bears that out, including Big-D Construction’s expansion of general-purpose robots across its jobsites.
On the infrastructure side, FieldAI has published work with NVIDIA Omniverse on the next generation of industrial AI, and a separate collaboration to accelerate adoption of AI among industrial customers. Simulation environments of that kind are how a robot-agnostic model gets exposure to conditions that would be slow, expensive or unsafe to gather on live sites.
The company is headquartered in Mission Viejo, California, and its stated aim is simply robot intelligence built for the real world.
