HUMANOIDS
The trial took place over six weeks at the start of the year, taking place both in the Humanoid Lab and later at the Cologne plant. The robot was given two core tasks: autonomously moving filled totes weighing 8kg between workstations and finding and manipulating a large, sheet metal car body component and moving it to a kitting table. Both tasks are commonplace in vehicle manufacturing, yet notoriously difficult to automate due to variability in objects, environments, Humanoid reported that in the first task, the robot operated continuously for a full hour, double the initial target, while achieving 97% reliability in autonomous pick-and-place operations. Productivity also surpassed projections, with the robot completing 83 picks per hour compared to the anticipated 50, a 60% improvement. The second workflow pushed the robot into even more complex territory. Here, Alpha demonstrated dual-arm manipulation, handling a thin sheet-metal car body component, an object that is both flexible and sensitive to improper handling. The robot successfully navigated to the storage location, and movement requirements.
identified and localised the part, lifted it securely, and placed it onto a kitting table. The key enabler, Sokolov adds, was the robot’s AI: “The system is built around task-level understanding, which allows the robot to adapt much faster. It combines perception, decision-making, and motion in real time, meaning we don’t need heavy re-engineering every time the robot encounters a new environment,” he says. Humanoid says technology from NVIDIA played a significant role in enabling this rapid development cycle. The project used operational digital twins, virtual replicas of real- world environments that allow teams to simulate, test, and refine robotic behaviour before physical deployment. What makes the tools such as NVIDIA Omniverse to create achievement more striking is the minimal time required to prepare the system. According to the company, just one hour of on-robot data collection was needed to fine-tune the AI model for the task. This rapid deployment was made possible by training the system on extensive datasets gathered across varied environments, allowing it to generalise
quickly to new conditions. “Because our models are trained on large, diverse datasets beforehand, we don’t need massive amounts of on-site data to deploy,” Sokolov says. “About an hour was enough to get a high-performing system. Simulation using tools like digital twins allows us to train and test the system before deployment, and minimal fine-tuning on-site makes the approach scalable across factories and industries.” And Humanoid’s work for Ford is just one of a number of collaborations taking place between humanoid robotics companies and automotive manufacturers at the moment. Elon Musk’s Tesla has said that it plans to get its Optimus robot models to help in its car plants from this year while Tesla has committed to converting space in its Freemont factory, previous used for Model S and X production to mass produce humanoids. Hyundai too has said it will deploy Atlas humanoid robots in its factories from 2028 following its investment in the robotics company behind Atlas – Boston Dynamics – in 2021. Not to be outdone, in February BMW announced a pilot to deploy humanoid
robots at its Leipzig plant in Germany in partnership with Hexagon Robotics following a previous demonstration at its Spartanburg plant in the USA in collaboration with tech firm Figure AI. “What differentiates us is speed and commercial focus. On the hardware side, we’re moving very fast. Our wheeled robot was built in about seven months, and our bipedal robot was developed even faster, in around five months. This level of speed is unprecedented for the industry,” Sokolov says. “But more importantly, we focus very early on real- world use cases. While most humanoid robotics companies focus more on impressive prototypes, we want to create robots that the market and different industries really need. That’s why we started testing with commercial
partners even at very early stages. In less than two years, we’ve already completed seven POCs, which gives us real feedback from the field. And then on the AI side, we recently introduced KinetIQ, our proprietary ‘AI brain’. It’s a four-layer framework that orchestrates entire robot fleets across industrial, service, and home environments, and a foundation of what we call a capability factory. “Our goal is very clear: to become a leading general- purpose humanoid robotics company for industrial applications in the next two years.” And beyond hardware and AI, Sokolov is also focused on the broader vision for humanoid robots in industrial environments. “We anticipate that in the near term, humanoid robots
will have the biggest impact in industrial environments. They will take on physically demanding, repetitive, or hazardous tasks, which will naturally increase productivity. Because robots don’t need rest or sleep, operations can run 24/7, unlocking a completely new level of efficiency,” he explains. “At a larger scale, thousands of robots working across factories and warehouses could free people from tasks that are repetitive or undesirable. It will open the door to much bigger opportunities, from continuous research to large-scale innovation, and eventually even areas like space exploration. At the core, it’s about shifting human effort toward higher-value work, while machines handle the rest.”
Written by Lucy Barnard, Editor, Automation News
20
21
Powered by FlippingBook