Period covered 19 – 25 October 2026Written by Tomo Miyata
This week I chose five stories about large companies investing in, and deploying, software that instructs robots from different makers and people.
My lead story is Körber's investment in NAiSE, a fellow German software company. Körber makes software for supply chains and technology for manufacturing. NAiSE's software runs automated guided vehicles (AGVs), autonomous mobile robots (AMRs) and human-driven industrial vehicles such as forklifts, whatever their maker, through a single system. An AGV follows a fixed route; an AMR chooses its own route. Logistics facilities are adding more of these vehicles, along with equipment that moves and stores goods (material handling equipment).
NAiSE says its software works whatever WMS, WES and WCS run behind it. A WMS manages inventory and shipping. A WES decides in what order and how incoming orders are assigned to work. A WCS tells the material handling equipment inside a facility which unit should move where next. Suppose company A's AMR, company B's AGV and a human-driven forklift share one aisle. Each maker's management software cannot see the other vehicles' positions, so they can end up deadlocked at an intersection, each yielding to the other.
Getting robots into a facility is getting easier. KION, a German maker of equipment for logistics facilities and factories, used AI to strengthen a digital twin, a virtual replica of a facility, at logistics group GXO's site in Épinoy, France. It cut the time to commission autonomous vehicles on site from weeks to days, and on-site work fell by more than 60%. Teradyne Robotics of the US, whose group includes robot-arm maker Universal Robots, partnered with Germany's Sereact. Sereact says one AI model can run robots without retraining for each new product. ABB of Switzerland added a mechanism that lets a robot that loads apparel onto a sorter relearn automatically from failures during operation, and says it can cut commissioning effort by up to 75%.
At the same time, I think the more kinds of equipment there are, the harder the work of the instructing side gets. An article on humanoid robots from Logistics Viewpoints says that even in facilities where humanoids work, WES, WMS and control software will coordinate the schedules of inventory, staff, equipment and transport. It also writes that the winner may be the company that fits robots into the workflow best, not the one that builds the best robot. The same article points out, however, that there are still no measured figures on the cost-effectiveness of humanoids.
I expect the robots themselves, and the AI models that drive them, to keep getting cheaper and easier to deploy. If so, the difference between facilities will come down to whether one piece of software can tell material handling equipment from different makers, and people, the order of work and where to go. A second difference is whether the operating company has drawn a line on how much the AI decides. For example, when a person steps in front of a robot carrying goods, should it stop or detour? The robot control system (RCS) carries out that action, but if nobody has decided the standard for it, then when an accident or delay occurs, it is unclear who is responsible.
Adding material handling equipment without a system that instructs it leaves each maker's vehicles isolated, goods waiting and stops increasing. Körber's own announcement says that without upper-level software controlling vehicles from several makers together, the fleets stay isolated and efficiency and scalability hit a ceiling. GROUND's product GWES is designed around a core function of instructing the material handling equipment and people in a logistics facility. It is not a combination of other vendors' software bolted on afterwards. Japanese logistics facilities face a serious labor shortage and handle a wide variety of goods in the same facility. I believe whether a facility has a system that instructs equipment and people will decide its productivity.
GROUND Inc. — Founder & CEOTomo Miyata
This Week's News
Five stories
01Lead storyEquipment inside a logistics facility (illustrative). Photo: Unsplash
InvestmentMulti-vendor fleet managementGermany
Körber — invests in Germany's NAiSE, which manages robots and vehicles from any maker together
Körber of Germany, which makes software for supply chains and technology for manufacturing, announced on October 22 that it will invest in NAiSE GmbH, also of Germany. The size of the investment and the stake were not disclosed. The two companies had already worked together, and the investment upgrades that to a strategic partnership. NAiSE was founded in 2017 and has operated without outside funding until now. Co-founders Kai Przybysz-Herz, Robert Libert and Raphael Kusumoto will continue to run it, and its existing partnerships are unchanged.
NAiSE's fleet management software runs AGVs, AMRs and human-driven industrial vehicles under one common control. The company's intralogistics platform connects vehicle management, vehicle traffic coordination, order management and inventory management, and does not depend on any maker. It has a standardized connection layer and, the company says, can be configured whichever WMS, WES and WCS run behind it.
The announcement says that as automation advances, operating companies end up using fleets from several makers, but without upper-level control each fleet stays isolated, which limits efficiency, visibility and scalability. The two companies will offer maker-independent vehicle management together with Körber's software for logistics facilities. NAiSE will open the AGV and AMR makers it works with to joint customers and new customers. They also plan to develop new uses, including humanoid robots, but gave no timing. Dr. Helena Garriga, a member of Körber's executive board responsible for the supply chain business, said, "We have gained a partner working on orchestrating vehicle fleets from different makers." Kai Przybysz-Herz, NAiSE's chief executive, said, "We deliberately chose a partner that respects our independence." NAiSE will keep its doors open to other system integrators and vehicle makers, he said.
KION — cuts on-site commissioning of autonomous vehicles from weeks to days, applying AI-based virtual pre-testing at a GXO facility in France
KION, a German global leader in industrial vehicles and logistics automation (its group includes automation systems company Dematic), presented results of physical AI in real operation on October 21 at NVIDIA's GTC Berlin 2026. Physical AI means AI that operates real equipment. At logistics group GXO's facility in Épinoy, France, KION's autonomous vehicles are used for daily work in the same space as people and other equipment. The trial run before deployment took place in early 2026.
The core of the results is the digital twin, a virtual replica of a real facility. Spatial data captured at the customer's site is added, and the vehicles' control software is trained, tested and tuned before deployment. According to KION, commissioning an autonomous vehicle fell from weeks to days, and on-site work at Épinoy fell by more than 60%. The mechanism for operating the vehicles safely near people was developed with NVIDIA's safety framework Halos. Consulting firm Accenture helped build the digital twin.
KION also presented RoCaP, a robotic case picker. An autonomous vehicle with a robot arm identifies cartons with an AI camera and picks and stacks them one at a time. It can be installed in existing facilities and handles cartons with damaged packaging. It was delivered in October 2026 to a facility of German retailer ROSSMANN. Because speed and gripping performance exceeded expectations in the trial, ROSSMANN ordered additional vehicles. The number was not disclosed.
Teradyne Robotics × Sereact — deploy robot arms that handle unfamiliar products with one AI model in Zalando's returns processing
Teradyne Robotics of the US, whose group includes robot-arm maker Universal Robots (UR) and mobile robot maker MiR, announced a partnership on October 21 with Sereact of Germany, which develops physical AI for logistics. Sereact's AI platform "Cortex" will operate UR robot arms to automate loosely defined tasks such as processing returns, picking single items and replenishing stores. Cortex is an AI model that outputs motions from images and language instructions. Sereact says one model runs all the robots without retraining for each new product or task.
Training uses NVIDIA's simulation environment Isaac Sim, and reinforcement learning uses Isaac Lab. In the robot cell (work area) on site, NVIDIA's Jetson Thor performs the motion inference. Teradyne Robotics will provide simulation data for UR and MiR equipment to NVIDIA's libraries.
Deployments include two facilities of European fashion retailer Zalando, operated by logistics group CEVA Logistics. In Greven, Germany, and Świebodzin, Poland, robots process returned apparel automatically. Installation at these two sites is the first stage in extending returns automation across Europe. Sereact says returning items quickly to a resalable state also helps meet the European Union rules that ban destroying returned goods. Hundreds of UR dual-arm systems have shipped to customer sites in Europe. Jacob Pascual Pape, Teradyne Robotics' chief commercial officer, said, "Physical AI only matters once it delivers results in real operation." Dr. Ralf Gulde, Sereact's chief executive, said, "One model runs all the robots."
ABB Robotics — adds AI that relearns from failures during operation to its apparel sorter-loading robot, cutting commissioning effort by up to 75%
ABB Robotics, the robotics division of Swiss industrial robot leader ABB, added "AI Robot Trainer" to "Fashion Inductor," a robot module that loads items onto an apparel sorter. ABB positions it as the first in its line of industrial physical AI that uses camera images. Fashion Inductor separates bagged garments and accessories into single items and loads them onto the sorter. Each cell handles up to 1,300 items an hour, with a picking accuracy target above 99.5%. Just Style reported this on October 22.
AI Robot Trainer supplies an image recognition model pretrained for apparel, plus tools for each use. This lets system integrators install and tune it without AI specialists. During operation it repeats the following cycle: it monitors robot performance, detects anomalies, generates new training data, retrains the model for that cell and returns the improved model to the robot. ABB says engineering and commissioning effort can fall by up to 75%.
A digital twin, a virtual copy of the equipment's software, links the control platform, path-planning software, AI Robot Trainer and the RobotStudio design tool to speed commissioning. Grippers (the parts that grasp goods) from other makers can also be integrated. ABB will extend the same approach to other camera-based uses by the first quarter of 2027. Customer names, prices and availability were not disclosed. Craig McDonnell, ABB Robotics' business line manager, said, "Physical AI fundamentally changes what robots can do in industrial applications."
Logistics Viewpoints — notes humanoids are starting to enter facilities, but there are still no measured figures on cost-effectiveness
Logistics Viewpoints, a US outlet carrying logistics analysis, published an article by Angeline Fernandes on October 19. It says humanoid robots have left laboratory demos and begun entering working facilities, but it has not yet been shown that they make business sense. Logistics group GXO has US maker Agility Robotics' bipedal robot Digit move totes (containers for goods) between automation equipment and conveyors. According to Agility Robotics, Digit has moved more than 100,000 totes at customer sites and logged more than 65,000 operating hours in total. The article does not verify these figures.
The humanoid tasks the article lists include moving totes, replenishing materials, feeding parts to production lines, assisting with trailer unloading and assisting at workstations. AMRs and automated storage systems that combine shelving with transporters, on the other hand, already have a track record and are well optimized. The article sees the humanoid's advantage as flexibility: one unit might do work that needed a combination of fixed equipment and people. Whether that flexibility pays for itself, it says, is unresolved.
The article does not give measured figures for cost or savings. As an example it cites work in which three people moved totes across several shifts. If a robot can do that reliably for long hours, labor cost falls and quality stabilizes. If uptime is unstable or much human supervision remains, the benefit disappears. The metrics to judge, it says, are throughput, uptime, maintenance, handling of exceptions and the cost of introduction. It also writes that coordinating a robot's work with the schedules of inventory, staff, equipment and transport, and with orders, requires WES, WMS and control software.
