INSAIT unveils SPEAR-1 – Europe’s first open robotic foundation model trained on 3D data

The model achieves state-of-the-art performance with up to 20 times less robotic data, marking a breakthrough in scalable and data-efficient robot learning.

INSAIT, part of Sofia University “St. Kliment Ohridski,” has announced SPEAR-1 – the first open robotic foundation model developed in Europe, achieving state-of-the-art performance with up to 20 times less robotic training data.

SPEAR-1 introduces a groundbreaking approach by learning from both robotic and non-robotic 3D data. This innovation marks a major step forward, as 3D data is abundant and easily accessible. With this method, SPEAR-1 matches or outperforms leading models such as OpenVLA, π0-FAST, and π0.5, all of which require 20 times more robot demonstration data.

This achievement represents a significant milestone for the next generation of intelligent robotic systems. Robotic foundation models are often described as the “ChatGPT equivalent” for robots – a single model capable of performing many tasks across diverse environments and robotic platforms. Traditionally, collecting sufficient robotic training data has been slow, costly, and complex. By leveraging the wealth of available 3D data, SPEAR-1 effectively addresses this bottleneck, paving the way for scalable and efficient robot learning.

SPEAR-1 is open-weight and general-purpose, capable of controlling a variety of robots through natural language instructions. This advancement opens the door to a new era of flexible, adaptive, and accessible robotic intelligence – one that will power robots in homes, factories, and cities.

The project was developed entirely by a team of INSAIT researchers: Nikolay Nikolov, Giuliano Albanese, Sombit Dey, Aleksandar Yanev, Prof. Luc Van Gool, Dr. Jan-Nico Zaech, and Dr. Danda Paudel.

More information and technical details are available on the official project website: spear.insait.ai