Field AI

Field AI develops AI software that enables autonomous operation of robots and vehicles in unstructured, unmapped environments without relying on GPS, prior maps, or human intervention. The company's core technology revolves around Field Foundation Models (FFMs), which allow machines to understand how to move in and interact with the physical world rather than just where to move.

Unlike traditional robotics solutions that require pre-mapped environments, Field AI's software enables robots to adapt to constantly changing conditions and new environments in real-time. The company's technology has demonstrated capability across various robot types including legged, wheeled, flying and tracked vehicles. Field AI's robots have completed tens of thousands of kilometers of fully autonomous inspection-focused traversal, collecting up to several terabytes of multimodal data per hour. The technology has proven particularly valuable for inspection tasks in industrial and construction environments where conditions continuously change.

Field AI's approach emphasizes environmental understanding over static mapping through heavy probabilistic modeling. Rather than requiring lengthy setup processes or human supervision, the company's robots can discover and navigate any environment autonomously with the press of a button. This capability stems from the team's experience in the DARPA Subterranean Challenge and DARPA RACER program, where their autonomous systems successfully navigated complex underground environments and off-road terrains without prior mapping.

Key customers and partnerships

The company has acquired commercial customers across industrial and construction sites worldwide where its robots perform autonomous inspection tasks. As of 2023, Field AI was operating its autonomous robots for paying customers at industrial and construction sites globally, where the systems can be deployed with minimal supervision for extended periods.

Field AI has also achieved significant results through the DARPA RACER program, becoming the first team to successfully complete all eight courses in their initial attempt in 2022, traversing several kilometers autonomously without GPS, prior maps, or existing trails.