fieldai raises $405m to build universal robot brains

FieldAI Secures $405 Million to Advance Embodied AI Robotics

FieldAI, a robotics startup headquartered in Irvine, California, has successfully raised $405 million through a series of funding rounds that were not previously publicized. These funds are dedicated to the development of what the company terms “foundational embodied AI models.” Essentially, these are advanced AI systems designed to function as the “brains” of robots, enabling adaptation across a wide range of platforms, including humanoids, quadrupeds, and autonomous vehicles.

Recent Funding and Backers

The company publicly announced the funding on Wednesday. The latest investment, totaling $314 million, was secured in August and was jointly led by Bezos Expeditions, Prysm, and Temasek. FieldAI also benefits from the support of prominent investors such as Khosla Ventures, Intel Capital, and Canaan Partners.

Embodied AI: A New Paradigm

Embodied AI distinguishes itself from conventional AI, which typically focuses on processing data like text or images. It centers on AI systems that directly control physical robots operating within real-world environments.

FieldAI is developing “Field Foundation Models,” which are versatile, physics-based embodied AI models. This methodology equips robots with the capacity for rapid learning and adaptation to novel environments, while simultaneously maintaining an awareness of potential risks, as explained by FieldAI’s founder and CEO, Ali Agha, in a recent TechCrunch interview.

The Vision: A Universal Robot Brain

“Our core objective is to create a unified AI brain capable of functioning across diverse robot types and a multitude of environments,” Agha stated. “Achieving this necessitates robust risk and safety management as robots navigate new surroundings. Historically, traditional robotics models and approaches have lacked the inherent ability to address these critical safety concerns.”

Integrating Physics for Safer Robotics

Agha emphasized that incorporating a layer of physics into these AI models is crucial for enabling robots to learn safely in unfamiliar environments. This integration provides robots with an additional source of information for decision-making – particularly valuable when encountering new situations – moving beyond the purely reactive approach of traditional Large Language Models (LLMs).

He further noted that while a degree of AI “hallucination” may be acceptable in certain contexts, it can be detrimental when robots operate in hazardous environments or interact with humans.

Confidence and Risk Thresholds

“This introduces a sense of self-awareness – understanding what the robot knows and, crucially, acknowledging when it lacks information or is uncertain about a decision,” Agha explained. “When the network gains access to this awareness, it begins to make significantly safer choices. It doesn’t simply output the next action; it also indicates its level of confidence, allowing customers to define acceptable risk thresholds, to which the robot will then respond.”

Decades of Research Culminate in FieldAI

Agha’s work on this concept spans decades of experience in roles at institutions such as NASA and the Massachusetts Institute of Technology (MIT). He launched FieldAI following a key technological advancement that enabled a single AI brain to effectively control various robot types, performing both identical and unique tasks.

Early Adoption and Future Growth

Since its inception in 2023, FieldAI has secured contracts across several industries, including construction, energy, and urban delivery. The company has chosen not to disclose the names of its current clients.

The newly acquired funding will be allocated to ongoing research and development efforts, as well as to scaling up production to meet customer demand and expand the company’s global presence.

Inspired by Evolution

Agha draws a parallel between FieldAI’s approach and the process of human evolution. “Humans have evolved to perform a wide range of tasks in diverse environments, possessing the ability to learn rapidly. We believe this is essential for robotics. While specialization for a single use case is possible, it is not the target market for our technology.”

We are committed to continuous improvement, and your feedback is invaluable. Please consider completing this survey to share your thoughts on TechCrunch’s coverage and events, and you’ll have a chance to win a prize!