Modern production systems that act independently and learn from their experiences are increasingly becoming the standard. This even goes so far as to suggest that humans will no longer be necessary on the store floor in the future - at least if the strategists have their way. What is certain, however, is that roles and responsibilities will shift. NTT Data presents the two major trends in manufacturing, examines their potential and answers the most important questions.

Trend 1: Humanoid robots

These machines, which are modeled on the human body in terms of shape and movement, represent a new stage in the development of automation: they not only carry out individual, predefined processes, but also make decisions independently and adapt to changing situations. This makes them fundamentally different from traditional industrial robots, which still rely heavily on manually programmed routines and only function reliably in strictly controlled environments.

This paradigm shift is made possible by the rapid progress of multimodal AI models that combine vision, hearing, speech and motor skills. These models can recognize objects, understand verbal instructions, plan complex sequences and break them down into precise subtasks. At the same time, the way in which robots acquire their skills is changing. Instead of exclusively learning in real test environments, a three-stage development model now forms the basis. First, large AI models analyze huge amounts of real and synthetic movement and sensor data. Basic rules for gripping, balancing or navigating are derived from this data. The systems then refine these skills in digital twins, which provide such precise physical simulations that even complex scenes such as climbing stairs, handling varying weights or reacting to unforeseen events can be trained realistically. The results are then transferred to on-board computers with sufficient computing power to link perception, planning and motor skills in real time. However, the current humanoid robots are limited in their possible applications. They can perform simple tasks such as gripping, sorting or moving materials. Complex assembly work, fine motor tasks or highly dynamic interactions with humans are still technologically challenging and require further progress in the areas of sensor technology, safety mechanisms and real-time planning.

Humanoid robots are coming into focus in areas where staff shortages, a high proportion of manual work steps or ergonomically demanding tasks hinder efficient operation. Demographic pressure is acting as an amplifier here: in almost all sectors, the demand for systems that take on tasks that were previously the exclusive preserve of humans is increasing. The spectrum of potential users is correspondingly broad. In industrial production, interest is growing, especially where high variance, tight spatial conditions or continuous material movements dominate. Projects in automotive production show that humanoid systems can perform logistics and material flow tasks almost autonomously. What is often forgotten: These robots can perform tasks that could previously only be carried out by trained specialists. Nevertheless, their human physique is not the right solution for all industrial applications. According to the "form follows function" principle, specialized systems or alternative automation concepts are often the better choice at many points in the production line.

Companies should therefore proceed as follows: First, manufacturers should analyze which processes can realistically be automated and which still require human expertise due to high fine motor skills or tight tolerances. On this basis, pilot projects with clearly defined success criteria should be selected. At the same time, it is crucial to harmonize the database. This ranges from movement and quality data to system statuses and real-time information from the material flow. Many production lines work with heterogeneous data formats, which encourages incorrect decisions or inefficient movement sequences. Robust IT/OT security is just as important. As humanoid robots operate in physical space, secure communication channels, segmented networks, clear approval mechanisms and transparent audit logs are essential. In view of the fact that the market is currently not very standardized, a modular architecture based on open protocols and interchangeable AI modules is also recommended in order to avoid dependencies. Last but not least, a KPI-based approach to cycle times, availability or error rates, for example, including a critical assessment of cost-effectiveness, is expedient to ensure that pilot projects do not fail due to exaggerated expectations.

Trend 2: Autonomous factories

Behind the vision of a completely autonomous factory, often referred to as a "dark factory", is a radical paradigm shift in industrial production. In theory, such a plant no longer needs light because it does not require human presence. In practice, this is not entirely true, as optical sensors and cameras are still dependent on lighting, but it aptly describes the target image: a production environment in which machines take over, coordinate and optimize all processes in real time. The key concept here is autonomy and not automation. While only sub-processes are automated in Industry 4.0 factories and employees are still required as decision-makers and problem solvers, autonomous production is gradually shifting these roles to intelligent systems.

Autonomous factories are made up of several closely interlinked layers. The Internet of Things (IoT) acts like a nervous system that permanently connects machines, workpieces, means of transportation and quality sensors. Thousands of signals flow into central systems without interruption. There, the AI takes on the role of the brain, so to speak: it recognizes patterns, predicts disruptions, adjusts production plans in the event of material bottlenecks and dynamically controls entire process chains. This approach is now being implemented on a large scale in Asia, in particular via so-called autonomous production twins. These are digital twins of the factory that not only simulate but also make decisions, for example if deliveries arrive late, machines break down or orders have to be prioritized at short notice. The result is significantly higher productivity, a lower error rate and reduced operating costs. However, this does not work without a far-reaching infrastructural transformation: a dark factory is not an upgrade of an existing Industry 4.0 production line, but a system that must be designed for digital decision-making logic throughout. Complex systems with years of history or proprietary control systems require an extremely high level of integration before autonomous behavior is even reliably possible.

Jochen Gemeinhardt is Head of Production & Supply Chain at NTT Data DACH: "Humanoid robots and autonomous factories are still more of a vision with initial pilot projects than a widespread reality. However, industrial value creation will change noticeably when these robots become the link between man and machine and factories control entire production systems independently. The advantages are obvious: overall equipment effectiveness will increase, as will quality and flexibility. This makes completely new business models possible. However, manufacturers should not underestimate the risks." © NTT Data

Autonomous factory concepts are no longer just relevant for car manufacturers. Rather, they are becoming increasingly important in all branches of industry in which recurring processes, high quality requirements and fluctuating market conditions come together. This is the case in mechanical and plant engineering, for example. The autonomous factory is therefore becoming a strategic instrument, particularly for globally active companies with a high number of variants, in order to secure delivery capability, margins and a competitive position in the long term.

The path to autonomous production is not a "big bang", but a transformation process lasting several years in which architecture, data flows, processes and governance are redefined. A methodical approach that realistically assesses technological maturity, integration costs and economic effects is crucial to success. As with humanoid robots, highly reliable and time-critical data on machine status, material flow, quality, energy consumption and logistical movements is the first prerequisite. This includes standardized data models for machines, systems and ERP/MES systems, edge architectures with a latency of less than 10 ms for critical processes and unique identities for workpieces and transport units. In terms of security, companies must segment their OT networks and implement zero-trust principles, introduce secure APIs between machines, MES, digital twins and AI platforms, establish monitoring for OT protocols and define governance rules. These determine when the AI is allowed to act autonomously and when human approval is required. One of the biggest risks is the fragmentation of factory IT. To avoid vendor lock-in, companies should rely on open communication standards (OPC UA, MQTT), interoperable digital twin models, containerized AI workloads and modular control technology. This is the only way to easily implement a change of provider, new functions or scaling steps at a later date. In short: autonomous factories cannot be realized at the push of a button. Even in highly digitalized production environments, mixed forms of automated, semi-autonomous and human-controlled processes will remain the norm for the time being.