automation solutions required individual programming of each robot. These robots were limited to simple, repetitive tasks, allowing for minimal flexibility. Training-based robotics: Embodied AI is a major technological breakthrough in robotics. The convergence of robotics, machine learning types such as reinforcement learning (RL), and advanced vision systems has transformed automation. By giving robots an understanding of the world, embodied AI enables new applications like bin picking. Unlike rule-based systems that depend on manual coding, AI-enabled, training-based robotics can now learn skills via RL in a trial-and-error approach in physical or simulated environments. Context-based robotics: Context-based, autonomous robotics are built on robotics foundation models (RFMs) and have a general understanding of the world.11 Because they require neither coding nor training by manufacturers, they can lead to a paradigm shift in robotics – that is, zero-shot learning. This considerably reduces the effort required to programme or teach these robots, opening the way to new, highly complex applications such as handling cables and addressing unforeseen events. RFMs are still in the development phase and are expected to break through in the coming years. Benefits and examples of the different robot types enabled by embodied AI TABLE 3 Robot type Benefits Example Rule-based robotics –Performance and reliability: Robots execute repetitive tasks with defined precision and speed, based on their coded robot programme.The vast majority of today’s global robot fleet is rule- based, including industrial robots in assembly lines and automated guided vehicles (AGVs) in logistics. For repetitive tasks that do not entail deviation or special requirements, such as automated assembly lines of medical products, rule-based robots will likely stay the norm in the future. Training-based robotics –Task versatility and flexibility: Robots understand the manufacturing world they have been trained in, giving them the necessary flexibility to adapt to different environments and the dexterity to handle known objects and perform versatile tasks. For example, a kitting robot can now handle a large variety of distinct parts, applying its learned skill library. –Situation adaptability: Robots can autonomously understand and solve unforeseen events in their trained domain by executing the required actions. For instance, if a screw gets stuck, robots can resolve the issue independently, significantly enhancing system reliability. Beyond articulated robots, embodied AI models enhance the capabilities of autonomous mobile robots (AMRs) for material transport drones and other robot types. –Manipulation dexterity: Robots can learn to move objects with dexterity, enabling them to conduct advanced movements such as contact-rich assembly of multiple gears, for example.Kitting robots in a warehouse can handle a diverse range of parts of varying dimensions and characteristics. Previously, kitting operations required manual intervention or multiple robots to manage different part sizes, lowering the return on investment. For example, Otto Group, an online retailer founded in Europe, has deployed AI-controlled robots in its fulfilment centre to handle the order-picking process. Thanks to its AI capabilities, the robot can process a wide variety of shapes, colours and quantities, which previously required human hand-eye coordination for items such as textiles.12 Frontier Technologies in Industrial Operations 16

Frontier Technologies in Industrial Operations 2025