Humanoid robots: teleoperation system enables human-like movements
A teleoperation system based on motion capture technology helps humanoid robots to make human-like movements. This works in real time.
The humanoid robot is controlled in real time using motion capture technology so that it can box like a human.
(Image: Yanjie Ze (Screenshot))
A team of scientists from Stanford University and Simon Fraser University has developed the teleoperated whole-body imitation system (Twist), with which humanoid robots can be controlled in real time with human-like movements. Twist combines motion capture technology with reinforcement learning and imitation learning methods.
“We want humanoids to have the same level of whole-body dexterity as humans,” says Yanjie Ze, first author of the study “TWIST: Teleoperated Whole-Body Imitation System”, which has been published in the preprint on Arxiv. “Imagine a messy kitchen. People can hold things with two hands and use their feet to move obstacles, such as a basket on the floor. Humans can also open the door with the sides of their body or their elbows. We want humanoids to be able to do the same by directly imitating humans.”
To do this, Twist uses data captured using motion capture technology with MoCap devices. The body movements of real people are precisely tracked and recorded using a MoCap suit with marking points and optical systems. Movement data is collected across all parts of the body – i.e., not only from arms, hands, legs and feet, but also from torso and hip movements as well as head movements. The data is converted into commands that humanoid robots can execute using artificial intelligence (AI).
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
The researchers used a humanoid G1 robot from Unitree for this purpose. They found that it was sufficient to capture whole-body movements and transfer them precisely to the robot's joints. At the same time, it is ensured that the movements of the various limbs are coordinated. The scientists also found that Twist offers greater accuracy in whole-body control than previous systems. Twist thus enables various movements and abilities in real time that were previously not possible. According to the researchers, Twist can also be used to control other humanoid robots.
The researchers emphasize that the full-body data collected is completely sufficient to control a humanoid robot. The movements are human-like, and the robots achieve full-body dexterity as a result. The captured data can also be used to train humanoid robots and teach them human-like movements and dexterity.
Videos by heise
The scientists now want to work on expanding the data acquisition. This should enable robots to learn autonomous skills.
(olb)
