RFM-1, introduced by Covariant, is a groundbreaking Robotics Foundation Model designed to provide robots with human-like reasoning capabilities. This 8 billion parameter transformer model is trained on a vast, multimodal dataset combining general internet data and specific real-world robotics interactions, enabling it to handle tasks across various modalities, including text, images, videos, robot actions, and sensor readings. RFM-1 excels in understanding physics through learned world models, allowing for precise prediction and simulation of physical interactions. Additionally, it leverages natural language processing for intuitive human-robot collaboration, significantly reducing the complexity of programming new robot behaviors. Read more