This Open Source Robot Brain Thinks in 3D

European roboticists today A powerful open source release Artificial intelligence The model that serves as the brain for art the robot— Helping them perceive and manage things with new skills.

new model, SPEAR-1Developed by researchers at the Institute of Computer Science, Artificial Intelligence and Technology (INSAIT) in Bulgaria This could help other researchers and startups develop and test smarter hardware for factories and warehouses.

Just as open-source language models have made it possible for researchers and organizations to experiment with generative AI, Martin Vechev, a computer scientist at INSIAT and ETH Zurich, said SPEAR-1 will help roboticists quickly test and iterate. “Open-weight models are critical to advancing embodied AI,” Vechev told Wired ahead of the release.

SPEAR-1 differs from existing robot foundation models in that it incorporates 3D data into its training mix. This gives the model an enhanced understanding of the physical world, making it easier to understand how objects move through physical space.

Robot foundation models are typically built on top of vision language models (VLMs) that have a broad but limited understanding of the physical world because the training comes from labeled 2D images. “Our approach addresses the mismatch between the knowledge of the 3D space in which the robot operates and the VLM that forms the core of the robotic foundation model,” said Vechev.

SPEAR-1 is roughly as capable as commercial base models designed for operating robots, when measured At RoboArena, a benchmark It tests the ability to get a robot to perform tasks such as squeezing a ketchup bottle, closing a drawer and piecing together key pieces of paper.

The race to make robots smarter is already on billion dollars Riding on it has spawned well-funded startups with commercial potential for generically capable robots. different And Generalist In addition to physical intelligence. The SPEAR-1 is almost as good as the Pi-0.5 in terms of physical intelligence. A billion dollar startup founded by an all-star team of robotics researchers.

SPEAR-1 suggests that the quest to build more intelligent robots could involve closed models like OpenAI, Google, and Anthropic, as well as open source variants like Llama, Dipsik, and Quen.

However, robot intelligence is still in its infancy. It is possible to train an AI model to operate a robot arm so that it can reliably pick specific objects from a table. However, in practice, the model must be retrained from scratch if a different type of robot arm is used or the object or environment is changed.