Innodata Opens New Jersey Lab for Robot Training Data

Innodata Inc. (Nasdaq: INOD) has opened a new research and development laboratory in New Jersey to generate data for training human-like robots. The facility in Ridgefield Park addresses an urgent need for high-quality training data for humanoids, industrial robots, and other forms of physical AI, expanding the company's existing physical AI practice.

Innodata's practice supports robotics programs across the full data lifecycle. This includes data collection, training data production, model evaluation, and safety assurance. The new lab is designed to compress development cycles and enable faster deployment of safer robots.

Sub-millimeter precision enables trustworthy validation

The New Jersey lab uses high-precision, low-latency infrared optical tracking cameras measuring movement down to the sub-millimeter level. Developed in collaboration with motion-capture leader Vicon, the system captures 3D data directly from bodies, whether human or mechanical. This method differs from providers that infer 3D motion from 2D video.

Vicon Managing Director Andrew Knox stated that their systems capture people, robots, and manipulated objects in the same space with sub-millimeter accuracy and millisecond latency. Vicon consulted on the lab's design and continues to provide technical support. The calibrated cameras observe robots externally, enabling comparison with a robot's own internal sensor telemetry. This provides independent measurements for validating performance data, testing claims, and establishing third-party benchmarks.

End-to-end capability accelerates robot development

Innodata CEO Rahul Singhal stated that physical AI is growing faster than any other segment in AI, but robotics teams face a lack of real-world interaction data that is expensive and slow to produce. The facility removes this bottleneck by offering a complete end-to-end capability from data collection through model evaluation.

The lab's experts can measure a robot's movements and responses in various scripted scenarios, including human-robot interactions. Customers can send in robots for evaluation to confirm they perform to specifications. The external cameras provide a check on a robot's often-noisy internal telemetry. Innodata applies rigorous, multi-stage quality processes honed over decades of delivering premium data for mission-critical use cases.

Customers gain access to usable, retargetable motion data

Innodata can produce training data for a wide range of physical AI platforms and retarget motion data from one platform to another. The practice supports human-operated hardware, wearable systems, sensor rigs, Universal Manipulator Interface grippers, and other multimodal capture setups. Customers can purchase motion-capture data in off-the-shelf packages or commission custom projects.

Franklin Tanner, vice president of robotics and physical AI at Innodata, explained the approach. "Physical AI has to earn its tokens one interaction at a time, and they have to be deliberate," he said. The company's sensors register the tiniest motion of every joint, which is critical for training a humanoid that can weigh almost 200 pounds.

Real-world data is considered the gold standard but is expensive and inefficient to collect at scale. Tanner noted that customers often request vast amounts of egocentric data, but a large percentage may be useless. The company is exploring ways to focus on higher-quality, useful data for Vision-Language-Action models with a university partner. Data can also help build digital twins to generate edge cases for training.

The laboratory operates as an independent robot evaluation environment. Innodata plans to make datasets generated by the facility available as prepackaged products for customers.