Toyota says letting factory workers train their own AI models saves more than 10,000 man-hours a year

Dec. 9, 2024 — Toyota said its in-house AI platform, which lets factory-floor employees build their own machine-learning models, has cut more than 10,000 man-hours a year from its manufacturing process. It is in use at all 10 of its car and unit factories. Here's what happened, what it cost, and what went wrong.

Cost and outcome (as of Dec. 9, 2024)

Outcome: Adoption "has allowed for a substantial reduction of over 10,000 man-hours per year in the actual manufacturing process," Toyota wrote. (Toyota on the Google Cloud blog, Dec. 9, 2024)
Cost: No total cost was given. Toyota said the hybrid design means it can "reduce the number of on-premises servers to one" rather than setting up GPU servers in every factory. (Toyota on the Google Cloud blog, Dec. 9, 2024)
Scale: Models created "rising from 8,000 in 2023 to 10,000 in 2024," with nearly 1,200 active users. (Toyota on the Google Cloud blog, Dec. 9, 2024)

The handoff it removed

Toyota said it began experimenting with AI in 2018, but "a shortage of employees with the expertise required for AI development created a bottleneck." In 2022 it set out "to democratize AI development within its factories." (Toyota on the Google Cloud blog, Dec. 9, 2024)

The goal was a platform "that enabled factory floor employees, regardless of their AI expertise, to create machine learning models with ease," automating manual tasks so workers could focus on process improvement. "We knew we needed to be led by those working on the manufacturing floor if we wanted them to use the AI more proactively," the team wrote. (Toyota on the Google Cloud blog, Dec. 9, 2024)

At the Takaoka factory, the platform inspects finished parts and is also used in the process itself: to "inspect the application of adhesives used to attach glass to back doors" and to detect abnormalities in injection molding machines for bumpers. (Toyota on the Google Cloud blog, Dec. 9, 2024)

How it was built

Toyota chose a hybrid of on-premises and cloud computing so it could "use on-premises resources during normal operations and scale to the cloud during peak demand, thus reducing GPU usage costs." It compared this to its own Just-in-Time method. (Toyota on the Google Cloud blog, Dec. 9, 2024)

"A small team of six developers" built and deployed the platform in about half the time of a standard Toyota system project; the section heading puts it at 1.5 years with six members. (Toyota on the Google Cloud blog, Dec. 9, 2024)

Creating a model takes from 10 to 15 minutes up to 10 hours. A change to how containers start cut model-creation time by 20%, which Toyota said improved the experience on the floor and led "to a surge in the number of users." (Toyota on the Google Cloud blog, Dec. 9, 2024)

More than 400 employees join in-house training programs each year, Toyota said. (Toyota on the Google Cloud blog, Dec. 9, 2024)

Toyota plans models with more detailed detection criteria, use in automated picking, and "maintenance and predictive management of the entire production line." It is considering generative AI to create test images where real defect data is scarce. (Toyota on the Google Cloud blog, Dec. 9, 2024)

What went wrong

Two different hours figures. The post says the platform "would be able to save us as many as 10,000 hours" and, later, that it achieved "over 10,000 man-hours per year." (Toyota on the Google Cloud blog, Dec. 9, 2024)
Vendor venue. The post was published on Google Cloud's blog and praises Google's services. (Toyota on the Google Cloud blog, Dec. 9, 2024)
No defect or cost data. Toyota gives hours saved and model counts, not defect rates or yen. (Toyota on the Google Cloud blog, Dec. 9, 2024)
What we don't know yet: how inspection accuracy compares with the manual checks it replaced.

Our follow-up question for Toyota

How do worker-built inspection models perform against manual inspection on defect escapes and false rejects?

Sources

Toyota Motor Corporation AI Group, “Toyota shifts into overdrive: Developing an AI platform for enhanced manufacturing efficiency,” Google Cloud Blog, Dec. 9, 2024. cloud.google.com

Disclosure: [not in copy] Editor placeholder