MANUS Gloves Now Natively Supported in NVIDIA Isaac Lab

MANUS gloves have been officially integrated into NVIDIA's Isaac Lab 2.3 as a native teleoperation device, enabling researchers and robotics teams to teleoperate simulated robots inside NVIDIA's Isaac Lab environment and capture high-fidelity demonstration data for robot policy training at scale.

The Data Quality Problem in Robot Learning

A sim-first approach to robot policy training streamlines development, reduces cost, and enables safer, more scalable deployment. But the approach is only as good as the data behind it. With Isaac Lab 2.3, NVIDIA has expanded teleoperation support to include MANUS gloves, making it easier to capture comprehensive, high-fidelity demonstration datasets directly inside simulation.

From Human Motion to Robot Policy

The MANUS glove data is streamed directly into Isaac Lab, which maps human hand configurations to robot hand joint positions in real time, enabling natural skill transfer from human to machine. For dexterous manipulation tasks, this places strict demands on the input device: tracking must be continuous, precise, and stable. MANUS gloves meet these requirements with:

• Millimeter-level precision across all finger joints, capturing the precise hand kinematics that dexterous manipulation policies depend on.
• Occlusion-free tracking with no reliance on external cameras or line-of-sight constraints.
• Drift-free data throughout extended operation sessions, ensuring data quality does not degrade over time.

Full Pipeline Integration

The MANUS x NVIDIA Isaac Lab integration supports the full dexterous teleoperation and data collection pipeline within Isaac Lab's workflow. Operators can record demonstrations for manipulation tasks, feed those demonstrations into Isaac Lab Mimic for augmentation and scaling, and use the resulting datasets to train policies via imitation learning—all within simulation, before any real-world deployment.

This positions MANUS gloves as a practical instrument at the foundation of the embodied AI development stack: a data quality guarantee for the demonstrations that policy training depends on.