Covariant, a world leader in AI robotics, has launched RFM-1, a Fundamental Robotics Model that endows robots with human-like reasoning capabilities. This release marks the first time a Generative AI model has succeeded in providing commercial robots with a deep understanding of language and the physical world, overcoming the reliability and flexibility challenges presented by traditional robotic systems based on manual programming or specialized learned models.
Since its founding in 2017, Covariant has been at the forefront of innovation in robotics , starting with picking and placing operations in warehouses. Combining a vast dataset of real robotic production data with an extensive collection of Internet data has enabled Covariant to unlock new levels of robotic productivity and envision applications in a wide range of industries.
Peter Chen, CEO and co-founder of Covariant, emphasizes the need for fundamental robotics models to access a large amount of high-quality multimodal data. The company has developed a highly scalable data collection system, collecting tens of millions of trajectories through a large fleet of warehouse automation robots deployed around the world.
Reasoning robots thanks to a Generative AI model
Covariant’s previous AI models have enabled robots to operate in a commercially meaningful way in a variety of operations and warehouse sectors. These robots have demonstrated the ability to adapt, understand complex scenes, handle objects never seen before and achieve levels of speed and reliability comparable to humans.
RFM-1, configured as an Any-to-any Multimodal Sequence Model, is an 8 billion parameter model trained on text, images, video, robot actions and physical measurements. This training allows the model to understand any modality as input and predict any modality as output, giving the robots a sophisticated ability to reason and make decisions in real time.
RFM-1’s specific capabilities include a physics model of the world that emerges from learning to generate videos, language-guided programming that facilitates robot-human collaboration, and learning from self-reflection that enables robots to rapidly improve their performance.
Not the end of the road
Pieter Abbeel, chief scientist and co-founder of Covariant notes that, despite recent advances in Generative AI in the creation of videos, these models are still disconnected from physical reality. RFM-1, trained on an extensive dataset rich in robotic physical interactions, represents a significant advance toward creating generalized AI models that can accurately simulate the physical world.
The introduction of RFM-1 by Covariant not only redefines the boundaries of what is possible in robotics, but also underscores the crucial role of Generative AI in bridging the digital and physical worlds. As these models continue to evolve, the potential for robotic applications in our daily lives expands, promising an era of unprecedented robotic autonomy.
Don’t miss any of our posts and follow us on social media!
Inspenet.com YouTube LinkedIn Facebook Instagram
Source: covariant.ai
Photo: shutterstock
Loading...
