AI for Robotics - NAVER LABS Europe

AI for Robotics

Robots are starting to make their way into our homes and offices but there’s a long way to go before they become sufficiently autonomous to be of real assistance in the constantly changing environments we live in.



One reason for the lack of autonomy of robots is that they simply do not have the ability to acquire and apply common sense knowledge or use it to accomplish tasks without detailed instructions. To leave a room, humans typically don’t need a map. Pinpointing where the door is situated is enough to know where to go. When we want to lift a glass of water, we do not need a detailed physical model describing the conservation of energy or the laws of gravity. In both cases, we use our common sense, developed from our past experience.

This capacity for common sense is what we are striving to teach robots, and we would like robots to acquire it not just from ‘past experience’, but also by taking appropriate exploratory actions. This capacity will allow robots to operate in real-world environments, performing useful tasks in collaboration with people.

To reach this goal, we’ve identified the following critical abilities a robot needs to acquire to move from highly-controlled environments into the real world.

  • Robust navigation and localization in dynamic environments with and without relying on precise maps or high-precision 3D perception.
  • Understanding the meaning of, and the relations between, the entities in the robot’s surrounding.
  • Learning to accomplish a task with minimal supervision from human operators.
  • Interaction with humans by e.g. understanding human language or gestures.

These abilities are possible with advances in AI, particularly in areas such as perception, weakly-supervised learning and reinforcement learning. Using these abilities, robots should be able to cope with an ever-changing world, navigate in dynamic environments, interact with us and even learn from us. They will quite simply become part of our everyday life.

The AI for Robotics theme comprises work from a number of our research areas.

A first-of-its-kind architecture that, based on a single image, predicts how a robot can pick up objects from within any scene could revolutionize applications in AR/VR and robotics. Blog article by Gregory Rogez
Open source release of the structure from motion and visual localization data format kapture. Blog article by Martin Humenberger
Dope featured image
A novel efficient model for whole-body 3D pose estimation (including bodies, hands and faces), that is trained by mimicking the output of hand-, body- and face-pose experts. Blog article by Philippe Weinzaepfel
ICRA blog banner
Our navigation system enables robots to adapt to specific real-world environments and use cases with only small amounts of human preference data. Blog article by Christopher Dance