Exploring the role of motion-capture in developing realistic robot behaviour #
Workers wearing full-body motion-capture suits are being filmed performing intimate human movements as part of an effort to train AI-powered robots, according to a video shared online by robotics company Somnia Robotics.
The footage, which has drawn attention on social media, shows how human body movements can be converted into digital data and used to develop more realistic robot behaviour.
The unusual training process matters because it highlights how rapidly robotics companies are moving beyond simple commands and repetitive industrial tasks. Researchers are increasingly relying on human demonstrations to teach machines how to move, balance and respond during close physical interactions. This is a technically difficult area for humanoid and companion robots.
Somnia's post has sparked interest because of the nature of the movements being recorded. While motion capture has long been used in films, video games and robotics, the footage has been interpreted online as showing the technology applied to intimate adult interactions. This has raised fresh questions about where the growing market for AI companion robots could lead.
How Motion-Capture Data Teaches Robots To Move #
Motion capture works by recording a person's body position and movement and converting them into digital data. That information can then be adapted for use by an animated character, computer model, or robot.
For humanoid robots, the process is more complicated than simply copying a human movement. A robot's joints, motors, weight, and physical limits differ from those of a person. Researchers must therefore retarget human motion data so that a machine can reproduce the intended movement without exceeding its mechanical capabilities. A 2025 study published in the Proceedings of the Conference on Robot Learning described motion capture as a promising data source for teaching humanoid robots through imitation. However, the researchers warned that movements physically possible for humans may not be directly achievable by machines with different body structures and force limits.
That challenge is especially important for movements involving two bodies in close contact, where balance, positioning and the ability to react to another person's movement all become more complex.
The Goal of More Natural Physical Interaction #
Somnia describes itself as a robotics and embodied AI company developing consumer companion robots. Its website says the company is working on lightweight robotic structures, motion control, tactile sensing and generative AI. Its first-generation companion robot remains in engineering development rather than on general sale.
The company says its robots are intended to move, sense touch and respond to people in close-range settings. Motion data gathered from human performers could help researchers develop smoother, more natural physical behaviour.
The broader robotics industry is also seeking ways to collect large amounts of real-world human movement data. Researchers behind the HumanPlus project, for example, developed a system that allows humanoid robots to follow human body and hand movements and use demonstrations to collect data for training autonomous skills.
Other projects have explored teaching robots from ordinary video footage rather than relying solely on specialised motion-capture systems. Researchers have demonstrated that robots can learn aspects of whole-body movement and manipulation by observing human demonstrations and adapting those actions to their own physical form.
What the Video Reveals About the Companion-Robot Market #
The Somnia footage stands out because intimate interactions are rarely discussed as openly as warehouse work, household chores, or factory tasks in mainstream robotics demonstrations.
Yet the underlying technology is part of a broader push to make robots capable of operating around people rather than simply performing pre-programmed actions in controlled environments.
Companies and researchers are exploring training methods that can help machines understand human movement and physical interaction across a wider range of situations.
A recent Reuters report noted that the robotics sector still faces a major shortage of real-world data needed to train increasingly capable embodied AI systems. Companies are investing heavily in collecting demonstrations that can help robots learn from human movements and perform physical tasks in real-world environments.
For now, Somnia's video offers a striking glimpse into one possible method of collecting that data. The company has shown people wearing motion-capture suits so robots can learn human positions, reflecting a broader approach in which human movement is used as training material for more physically capable robotic systems.
© Copyright IBTimes 2026. All rights reserved.