logo

NAVER LABS Europe Unveils DIVINE, a Lightweight, High-Speed Next-Generation “Robot Brain,” to Make AI Robots More Practical

- Universal encoder integrates diverse AI tasks, recognizing everything from 2D images to 3D spaces and people in a single process

- Distills the strengths of each model to ▲reduce encoder memory use by 90% and ▲improve system processing speed by up to four times

- Makes high-performance AI possible in compact robots, lowering barriers to AI robot adoption in everyday and industrial settings

- Related research accepted at leading global conferences CVPR and ECCV, demonstrating leadership in robot foundation models, a core technology for Physical AI

June 23, 2026

NAVER LABS Europe announced on June 23 the release of DIVINE, a universal encoder designed to help autonomous robots perform tasks more efficiently in industrial and everyday environments.

To understand their surroundings and carry out a range of tasks, autonomous robots have traditionally used several AI encoders at once. An encoder is a component that turns data gathered from sensors such as cameras and LiDAR into a format that AI models can process.

DIVINE brings these encoders together in a single universal encoder, supporting a wide range of visual AI functions, from image understanding to spatial and human perception.

Until now, tasks such as ▲localization, ▲depth estimation, ▲spatial understanding, and ▲human perception have each required separate AI models and encoders. As a result, robots had to process the same input data repeatedly, driving up memory use and computational load.

NAVER LABS Europe addressed this problem using multi-teacher distillation, a technique that distills what specialized encoders have learned and combines it into a single encoder.

The technique transfers only the essential knowledge from expert “teacher” models specialized in areas such as image understanding, spatial perception, and human perception to one “student” model. This allows the student model to handle a wide range of tasks without relying on multiple large expert models.

With DIVINE, NAVER LABS Europe has condensed the functions of multiple specialized encoders—for 2D image understanding, 3D spatial reconstruction, and human perception—into a single system.

This allows robots to perform a wide range of AI tasks with DIVINE alone, without relying on multiple separate encoders.

This matters most in settings where people and robots share the same space and robots must read their surroundings quickly and respond in real time. By handling multiple AI tasks through a single encoder, DIVINE helps robots interpret their environment quickly even with limited computing resources.

In testing, DIVINE reduced the computational burden while maximizing performance. Compared with multi-encoder systems, it cut encoder memory use by around 90% and made encoding up to 12 times faster. It also reduced overall robot memory use by about 62% and improved system processing speed by up to four times.

AI models for robots have traditionally depended on servers or high-performance computing equipment because of the massive amount of computation they require. DIVINE can run these same AI functions with less memory and computing power, making them more practical for onboard use. This is expected to help extend high-performance AI to a wider variety of robots.

In other words, robots equipped with DIVINE as a “fast, intelligent brain” can understand situations on their own and perform diverse tasks as autonomous AI robots, without requiring bulky hardware or expensive, heavy computing equipment.

DIVINE is also designed so new AI functions can be added easily. When AI models are upgraded, existing robots can improve their performance simply by updating DIVINE, without requiring the introduction of new robots equipped with the latest model.

“As Physical AI moves closer to commercialization worldwide, making the robot brain lighter has become a key priority,” said Lee Dong-hwan, Leader of the Vision Group at NAVER LABS. “DIVINE will help lower the barriers to adopting AI robots across everyday and industrial settings.”

Two NAVER LABS Europe studies related to DIVINE were also accepted by the European Conference on Computer Vision (ECCV) in 2024 and the Conference on Computer Vision and Pattern Recognition (CVPR) in 2025, respectively, demonstrating the company’s world-class technical capabilities. (End)