The video is a publicly available reference from YouTube, showcasing how the robotics industry is experimenting with real-to-sim-to-real training environments. While this video is not directly related to the use case described below, it offers valuable inspiration for those exploring similar concepts.
A world leading robotics company has introduced an advanced physical AI training facility designed to bridge the gap between AI models trained in simulation and their performance in real-world environments.
While simulation offers scale and speed, AI models often fall short without being anchored in real-world interaction. To overcome this challenge, this company combines simulation with structured, repeatable real-world robot training.
Although simulation enables large-scale training, AI models trained in virtual environments often struggle to adapt to the unpredictable complexities of the real world, also known as the sim-to-real (S2R) gap.
Traditionally, improving model performance requires collecting real-world data, a process that is both time-consuming and costly, especially when repeated for every new task or robot platform.
To address these limitations, the team developed a continuous real-to-sim-to-real training loop, combining synthetic training with live robot demonstrations under human teleoperation.
In this approach, AI models are first trained on synthetic data and then refined with real-world trajectories collected from physical robots performing tasks under human control.
During teleoperation, MANUS Metagloves Pro enable millimeter-level precision and low-latency human control of robotic hands. This allows operators to guide robots through dexterous tasks, generating high-quality real-world data which strengthens the training process.
All data generated during the training process is stored in a centralized cloud-based skill repository, allowing learned capabilities to be shared across different robot platforms and locations.
This creates a scalable infrastructure for physical AI, one that unifies simulation and real-world learning into an iterative, self-improving loop.
By integrating teleoperation with MANUS Metagloves Pro, this workflow shortens development cycles, improves robustness and adaptability of robotic skills, and accelerates the transfer of dexterous manipulation from labs to real-world environments.
