DeepHow brings physical AI to the manual work inside factories
Yazaki will deploy video-based workflow analytics to measure manual production work and identify bottlenecks across assembly lines.
DeepHow is bringing vision-language AI into a part of manufacturing that traditional industrial software often struggles to measure: the manual work performed by people on production lines.
What happened
Automotive-parts manufacturer Yazaki North America agreed to deploy DeepHow’s Time and Motion AI across production lines, beginning with operations in Mexico.
The system uses video and vision-language models to identify workflow stages, measure cycle times and compare how operators and stations perform. Yazaki expects the deployment to help identify bottlenecks and improve line balancing.
DeepHow says its platform serves more than 100 customers across 1,500 locations and can reduce analysis that previously took weeks to days or minutes. Those performance figures are company-supplied, and financial terms of the Yazaki agreement were not disclosed.
Why it matters
Factories have extensive sensor data for machines, but manual assembly work is harder to analyse systematically. Engineers often rely on time studies, observation and spreadsheets to understand where processes slow down.
Converting video into structured operational data could make those workflows easier to measure and improve.
The bigger picture
This is an example of physical AI being applied to existing industrial operations rather than controlling a robot directly. The technology could support training, process design and productivity analysis, but it also creates worker-privacy and surveillance concerns.
DeepHow and Yazaki will need clear governance over how video is collected, stored and used. The commercial value will ultimately depend on independently measurable improvements in throughput, quality or labour efficiency—not the speed of the analysis alone.