AI Unlocks Automation For Manufacturers Large and Small

Standard Bots’ cobots incorporate AI-powered features such as Click Find, which enables users to train a robot to recognize and locate object types by clicking on them in the camera’s live feed, and Text Find which enables users to type a natural language description of a part for the robot to find. Photos by Julia Hider for MMS.

AI might be the latest technology in manufacturing, but that doesn’t mean it’s just for the largest, most advanced factories. “You can take it for as small or as large as you want,” said FANUC America President and CEO Michael Cicco during the day one keynote at Automate 2026. “You can start to compile all of the NDS data in an entire NASA factory, and with the compute power that's available today, you contribute the biggest brain you can possibly imagine to do that, or you can have a very small or medium-sized business where all you're really trying to do is be able to reprogram that robot very easily every day, because your product changes every day.”

Ultimately, AI “is lowering the area of entry for small and medium-sized businesses,” Cicco continued. This will be the key to reshoring manufacturing, says Evan Beard, CEO of American robot manufacturer Standard Bots. In the day three keynote, he explained that China’s secret to manufacturing dominance is automation rather than cheap labor. China has installed more robots than every other country combined, while just 6% of American manufacturers have a robot. According to Beard, this adoption figure comes down to two challenges: robots are “too hard to use and not capable enough.”

Featured Content

Metalworking Fluid Tackles the Challenges of Titanium Machining READ MORE

35.BI-MU: Plenty of New Features for an Edition Rich in Thematic Insights READ MORE

WEBINAR: From Machine Data to Guided Action: How Modern Shops Are Closing the Execution Gap READ MORE

Standard Bots’ robots incorporate AI-powered features to tackle these issues. Click Find enables users to train a robot to recognize and locate object types by clicking on them in the camera’s live feed. A foundation model pretrained on billions of images speeds up the learning process. Text Find, built on the same foundation model, enables users to type a natural language description of a part (for example, “metal ring”) for the robot to find.

This palletizing demo from FANUC uses ROS 2 and NVIDIA’s Jetson Thor edge AI computer, dynamically altering the path of the cobot to avoid obstructions as it completes a palletizing task.

AI Alliances

AI is also shaking up how longtime robot manufacturers are operating. In the months leading up to the show, FANUC, which has traditionally operated with a more closed ecosystem, announced several releases and partnerships related to physical AI, starting with the availability of the open-source ROS 2 driver for its CRX-series cobots on FANUC’s GitHub in late 2025. ROS is an industry-standard open platform that enables different robot components and software to communicate and coordinate. This means users aren’t locked into FANUC’s proprietary programming environment and are now free to use open-source navigation, vision and planning tools alongside more components from different vendors.

In March, FANUC announced a collaboration with Nvidia which provides access to Nvidia’s Jetson edge modules, cloud/edge AI infrastructure, Isaac Sim open robotic simulation framework and Omniverse libraries, and more with its robots and Roboguide simulation software. This enables users to create digital twins of their factories, train robots virtually and deploy and redeploy them quickly.

In May, the company announced a partnership with Google to further advance its physical AI offerings. Google contributes to and helps maintain ROS through its AI robotics and software company, Intrinsic, and the collaboration helped deepen FANUC’s commitment to ROS.

The company’s Automate booth served as a showcase of physical AI applications enabled by these new collaborations. Demonstrations included a CRX-20iA cobot tightening bolts on a moving engine block, using technology from Inbolt and NVIDIA to track the part’s movement on a bidirectional conveyor belt for nonstop assembly; an M-710 series robot performing a palletizing task with the ability to move around obstructions, simulated and programmed with ROS 2 and Python; and a CRX cobot programmed using natural language commands.

I tested the UR AI Trainer from Universal Robots, which enables users to guide cobots through tasks, gathering high-fidelity data to train a custom AI model.

Custom AI

New technology is also making AI more accessible to users. Universal Robots was allowing attendees to test out its UR AI Trainer, which enables users to collect high-fidelity data to train physical AI models. One of the biggest challenges in physical AI is collecting high-quality training data that reflects the actual tasks and environments in which the robots operate. LLMs scraped the whole internet to train their models, but that’s just not possible for physical AI because the data doesn’t exist yet and still needs to be created. UR’s solution enables operators to guide UR cobots through tasks while a second robot or set of robots follows along in real time. Force feedback enables operators to feel contact as they guide the robots. Scale AI collects the vision, robot state, motion and force data and provides the infrastructure to create a custom physical AI model.

For quality applications, Siemens was demonstrating its Inspekto AI-powered quality system. Users can train it in as little as 30 minutes with as few as 20 good sample parts and no need for samples of defects.