MANY MINDEDInnodata opens motion-capture lab to train robots for natural movement
Innodata established a motion-capture laboratory in Ridgefield Park, New Jersey on September 30, 2026. The facility uses Vicon infrared optical tracking cameras to capture 3D motion data from human or mechanical subjects at sub-millimeter accuracy. This lab specifically targets 200 lb (90.7 kg) humanoid robots, enabling precise movements like safely handling objects where context matters—such as picking up a mug by the handle to avoid burns. Innodata states the lab generates high-quality training data to address scarcity in physical AI, allowing robots to validate performance against customer specifications through scripted human-robot interaction scenarios.
The lab’s mechanism tackles a critical bottleneck: generating realistic training data for robots that move naturally in real-world settings. By capturing detailed motion patterns with high precision, it helps robots learn context-sensitive actions—like object handling—without relying on incomplete or artificial datasets. This directly supports the development of assistive robotics that can operate safely in environments where human-robot interaction is necessary.
For mobility, this advances accessibility for assistive tasks by making robots more capable of nuanced physical interactions. If scaled, it could reduce the need for expensive, specialized human oversight in robot training—potentially lowering costs for robots that help with daily tasks. However, the lab currently faces challenges with complex scenarios (e.g., mugs on conveyors with varying handle orientations) and incomplete human responses during data collection, which may cause premature shutdowns. It also focuses primarily on single-agent interactions while working toward multi-agent systems.
*Note: The source provides limited detail on real-world implementation scale and long-term cost impacts. The lab’s current scope targets specific robot weights and interaction types, with caveats about data collection expenses and human response limitations.*
Source: The Robot Report