32 n ROBOTICS AND AUTOMATION July/August 2024 www.drivesncontrols.com Using AI to program robots and distinguish products Industry has been researching the use of AI (arti cial intelligence) for some time. One promising application is to use the technology to program industrial robots. “To a large extent, programming a robot for a task is still reserved for experts,” explains Roland Ritter, who manages Kuka’s simulation programme. “This is precisely why we are working on an AI chatbot that translates a simple voice command into programming code.” The AI model generates the code that makes the robot perform a sequence of actions such as: “Grab the components one by one, and place them in a U-shape on a table”. At the moment, Kuka is still using AI in a simulated environment to perform such tasks. “We could transfer the AI-generated code to the robot controller, but that is currently still too unsafe,” Ritter acknowledges. “The entire industry agrees on this.” The robot's digital twin steps in to check whether the AIgenerated robot program is error-free. Kuka has been training its AI chatbot with the extensive data available from its in-house programming language KRL (Kuka Robot Language), which it has been using for decades. In the next step, Kuka is working on transferring the AI chatbot to customers to gain experience in how they will formulate voice commands for robots. “AI models do not necessarily get better with more training,” Ritter points out. “It is important to nd the right balance so that the training data remains at a high quality level.” Kuka has also been working with its Swissbased intralogistics subsidiary, Swisslog, to develop an AI model that will be able to distinguish, for example, bottles of shampoo from shower gel, and to recognise waste materials in warehousing applications. Swisslog’s customers include large companies in the food and pharmaceutical sectors. These companies typically have tens of thousands of dierent products, that may be packed into bags or boxes – or distributed without any outer packaging at all. “Every day, these dierent items have to be picked – put together for a customer or delivery order – as error-free as possible,” explains Niklas Goddemeier, head of R&D at Swisslog’s Robogistic product centre. “To ensure that the individual products are put together correctly and that no leftover packaging nds its way to the customer, we have trained an AI model that can not only recognise waste, but can also distinguish shampoo bottles from shower gel bottles.” Swisslog has developed an AI- and camerasupported item-picking robot called ItemPiQ, which picks items from source bins and places them into order bins. It combines a seven-axis arm, a multifunction gripper and an intelligent vision system. The robot, which has a 1.1m reach, can pick up to 1,000 items per hour, depending on their characteristics, and can change its gripper autonomously to adapt to dierent products and types of packaging. The company has been using AI models to improve the gripping quality of this system. The multifunction gripper has a central suction cup supported by three ngers, with smaller cups surrounding it. The central cup can work alone or in combination with the ngers, allowing the ItemPiQ to pick a wide range of product shapes and sizes. The gripper learns as it picks. The rst time it encounters an item it will decide the best way to pick it. The next time it sees the same item, it will remember whether that choice worked and make improvements to cut cycle times and boost picking success rates. Swisslog says that AI is a natural choice for image-based robot systems. But it still needs to answer the question: how do such systems continue to learn? This topic of “model updates” is keeping Swisslog’s developers busy. n The robot-maker Kuka and its logistics subsidiary Swisslog are using AI modelling to program robots and to distinguish items in warehouses. They are also tackling issues such whether AI models improve with training. Kuka is using articial intelligence to help program its robots
RkJQdWJsaXNoZXIy MjQ0NzM=