Content |
Chronicle
2025: Robotics breakthrough: Robots now learn to move by watching themselves in video
At the end of February 2025, American specialists from the Columbia University School of Engineering announced the development of a system that allows robots to acquire movement skills by watching themselves on video. This eliminates the need to use complex virtual simulators.
As the lead author of the study, Yuhan Hu, notes, the goal of the project is to create a robot that understands its own body, adapts to damage and learns new actions without constant human programming. In most cases, robots first learn to move in special simulators. After acquiring certain skills in such a virtual environment, the robot is released into the physical world, where it can continue to learn. The better and more realistic the simulator, the faster the training process goes. However, creating a good simulator is a laborious process that requires the involvement of highly qualified developers. Hu's team proposed an alternative way, teaching the robot how to create a simulator of itself through observing its own movements through the camera.
 | Just as humans learn to dance by observing their reflection in the mirror, robots use raw video to form kinematic self-awareness, says Hu. |  |
Scientists have developed a platform by which robots can autonomously model their own 3D shapes using a single 2D camera. For this, three artificial intelligence systems are used to simulate the work of the brain. The technology allows the robot to understand its movements and optimize them. The new system also helps robots adjust movements, for example, in the event of limb damage or malfunction of the manipulator.[1]
2024: Human-free robot training system launched
In late November 2024, scientists at the University of California, Berkeley, unveiled a system to allow robots to exchange skills without human input. The RoVi-Aug platform generates synthetic visual demonstrations using different types of robots and shooting angles, increasing the versatility of learning. Read more here.
2019: Facebook launches open AI simulator to train robots in realistic environment
In mid-June 2019, Facebook released the AI Habitat simulator (along with the source code), which is designed to train robots so that they can navigate realistic conditions. Read more here.
Notes
| The site content is translated by machine translation software powered by PROMT. The machine-translated articles are not always perfect and may contain errors in vocabulary, syntax or grammar. Read original article If you find inaccuracies or errors in the results of machine translation, please write to editor@tadviser.ru. We will make every effort to correct them as soon as possible. |