Robot Retasking in High-Mix Manufacturing with Workr
Workr
Workr, a manufacturing AI company, uses NVIDIA Omniverse, Isaac Sim and accelerated computing to let on-site operators retask industrial robots in under five minutes via a tablet interface, eliminating weeks of traditional programming. The edge AI pipeline runs on 2x NVIDIA RTX PRO 6000 Blackwell Max-Q GPUs attached to the robot cell, using models including RAFT-Stereo, Detectron2, NVIDIA Isaac FoundationPose and NVIDIA Isaac cuMotion, trained with synthetic data generated in Isaac Sim digital twins. Deployed with customers Yuasa International and Haas Alfex.
Overview
Workr, a manufacturing AI company, uses NVIDIA Omniverse, Isaac Sim and accelerated computing to let on-site operators retask industrial robots in under five minutes via a tablet interface, eliminating weeks of traditional programming. The edge AI pipeline runs on 2x NVIDIA RTX PRO 6000 Blackwell Max-Q GPUs attached to the robot cell, using models including RAFT-Stereo, Detectron2, NVIDIA Isaac FoundationPose and NVIDIA Isaac cuMotion, trained with synthetic data generated in Isaac Sim digital twins. Deployed with customers Yuasa International and Haas Alfex.
The challenge
Industrial robots have existed since the 1960s, yet adoption outside high-volume production lines remains limited — almost 90% of manufacturing still remains unautomated in 2025. The challenge isn't robot capability, it's the prohibitive time and expertise required to deploy them: each new part or cell layout requires expert programming, calibration and fixture design that can take weeks or even months, so in high-mix environments the upfront effort often outweighs the benefit.
The solution
Workr built a software solution that lets on-site operators retask industrial robots in under five minutes using a tablet, via a five-step workflow (scan the part, set infeed/outfeed positions, place an example part, hit Learn). The entire AI training and inference pipeline runs at the edge on 2x NVIDIA RTX PRO 6000 Blackwell Max-Q GPUs attached to the robot cell. The pipeline uses a customized RAFT-Stereo variant for depth-map creation, Detectron2 with per-instance finetuned R-CNN models for segmentation, NVIDIA Isaac FoundationPose (optimized with NVIDIA TensorRT) for 6-DOF pose estimation, and NVIDIA Isaac cuMotion for CUDA-accelerated motion planning. Models are trained using synthetic, physically accurate data generated from digital twins of each robot cell built in NVIDIA Isaac Sim on NVIDIA Omniverse, refined further through human-in-the-loop correction on a tablet UI using the Segment Anything 2 model.
Reported business value
By running the entire AI training and inference pipeline on NVIDIA GPUs at the edge, new parts can be introduced and robot tasks deployed in under five minutes, dramatically reducing the downtime and engineering costs that have historically prevented automation adoption. The approach is deployed with customers Yuasa International and Haas Alfex.
Sources
Open any source and check the claim yourself — that is the point of the register.
Other manufacturing entries in the register.
ArcelorMittal Enhances Steel Production Through Digital Innovation
ArcelorMittal partnered with IBM Consulting and Infosys to modernize operations through AI, cloud technology and SAP S/4HANA migration. At ArcelorMittal Eisenhüttenstadt, machine learning was introduced to predict and prevent surface defects on automotive steel sheets. At the Hamburg wire rod plant, AI optimized the trimming process by analyzing historical production data to determine optimal cutting points, reducing trim scrap by 20% and contributing to energy savings and lower CO2 emissions. ArcelorMittal also deployed a bio-inspired Ant Colony Optimization algorithm to calculate optimal production schedules, reducing downtime and material waste. At AM/NS India, IBM Consulting used IBM Rapid Move for SAP S/4HANA to migrate data and applications from outdated platforms to a single SAP instance across locations in Dubai, Indonesia and India.
Michelin runs 200+ AI use cases across manufacturing, supply chain and innovation
French tire manufacturer Michelin has more than 200 AI use cases in production, led by group chief data and AI officer Ambica Rajagopal. Its in-house IRIS system, protected by over 20 patents, partially automates end-of-line visual tire defect inspection to improve inspector efficiency and workplace ergonomics while operators retain final accountability. Machine learning forecasting tools improve demand forecast accuracy and proactively detect stock shortages in the supply chain. Michelin scans the startup ecosystem and uses tools including Databricks and Dataiku, and has partnerships with Microsoft and Rockwell Automation to codevelop AI solutions. The company reports AI-project ROI exceeding €50 million per year, growing 30-40% annually for three consecutive years, governed by an internal data office and responsible-AI principles (people-centric, explainable, accountable).
Schneider Electric fast-tracks innovation with Azure OpenAI Service
Schneider Electric bases customer-facing AI solutions on Azure OpenAI Service within Microsoft Cloud for Manufacturing. Its EcoStruxure Microgrid Advisor uses Azure OpenAI Service and Azure IoT for dynamic control of facility energy performance, EcoStruxure Resource Advisor Copilot helps customers manage energy usage, and the company is developing a PLC code generation copilot to automate programmable logic controller programming for manufacturing robots and IoT devices.
Foxconn Develops Physical AI-Enabled Smart Factories With Digital Twins
Foxconn (Hon Hai Technology Group) uses physically accurate digital twins integrating NVIDIA Omniverse libraries and OpenUSD to design, deploy, and manage high-volume production facilities, including those producing NVIDIA GB200 Grace Blackwell Superchip systems. Its Fii Omniverse Digital Twin (FODT) platform creates virtual replicas of factories, enabling simulation-driven design, real-time monitoring, and optimized operations. Using NVIDIA PhysicsNeMo AI models, Foxconn achieves 150x faster computational fluid dynamics simulations for thermal analysis (minutes vs. hours). Standardized OpenUSD-based digital twin assets enable rapid migration of entire production lines between global factories (e.g., Taiwan to Mexico). Robot workcells and AGV logistics are simulated in FODT before physical deployment: complex robotic tasks such as screw tightening and cable insertion are simulated and refined with NVIDIA Isaac Sim, Isaac Lab, FoundationPose models, and NVIDIA cuMotion, while AGV path is optimized by connecting simulations with Material Control Systems and NVIDIA cuOpt. Foxconn also built video analytics AI agents with NVIDIA Metropolis and the NVIDIA AI Blueprint for video search and summarization to monitor factory floors, and developed FoxBrain, an AI platform powered by NVIDIA NeMo trained in four weeks, described as Taiwan's first large language model with advanced reasoning capabilities. Leo Guo, General Manager of Fii Robotic Group, said the company believes it can cut factory setup and planning time by about 50%.
Was this helpful?
Your feedback helps us improve our use case database