Innodata Opens Motion-Capture Research Lab for Humanoids in Collaboration With Vicon

Innodata Inc. said September 30 that it had opened a new R&D lab to generate training data for humanoids and industrial robots and independently verify performance data produced internally by robots. The facility is in New Jersey and was designed in collaboration with Vicon. The lab will sell motion-capture data, take on commissioned projects and evaluate robots.

Innodata said in its announcement that it will provide physical AI companies with an integrated offering spanning data collection through model evaluation. The approach puts the acquisition of training data, often a bottleneck in robot development, and performance verification into the same workflow. It is relevant to companies seeking to shorten development processes. The design also uses Vicon's optical measurement technology to compare robots' internal telemetry with external observations, making it applicable to verifying performance claims and checking safety.

The new aspect of the announcement is that Innodata has integrated physical AI services spanning data generation and performance verification. Rather than simply installing motion-capture equipment, the lab will handle 3D data capture, the commercialization of training data and third-party robot evaluations under one roof. This extends the company's data business while representing a substantial expansion into physical AI. As shown by related articles in this newspaper, Flourish One to Begin Sales of Home Robot and NVIDIA Announces Safety Management Platform, key issues surrounding humanoids include not only manufacturing the robots themselves but also combining training data with safety assurance. Innodata's facility focuses on the foundation for the former: how to collect and verify real-world movements. This could also help establish standards for measuring robot performance externally. The lab links the acquisition of input data with the conversion of 3D motion data for model training and verification before shipment or evaluation. Vicon's involvement in particular means the setup is designed to compare external measurements with robots' own sensor readings. In physical AI, the precision with which movements are recorded, as well as the volume of data, is seen as directly affecting performance. Submillimeter accuracy and millisecond latency could therefore be practical differentiators. For now, what is known is the lab's functional design; its capacity to handle projects and the scale at which it can supply data remain unclear. It is also unknown which robot types or applications will account for the greatest demand for evaluation services. Key questions going forward include the share of commissioned work, how standardized the data packages will be, the scope of evaluation methods and the division of responsibilities with Vicon.

According to the announcement, the lab is equipped with high-precision, low-latency infrared optical tracking cameras that measure movements to submillimeter levels. Innodata said it directly captures 3D data from the body of a person or machine, rather than estimating 3D motion from 2D video analysis. The company said customers can buy ready-made motion-capture data packages or commission custom projects. They can also bring in robots for evaluation. Comparing internal telemetry with external observations can be used to benchmark performance claims, reliability and safety, the company said. Evaluations will also cover scripted scenarios in which people and robots interact.