Commentary: Four trends for physical AI in 2026

1. predictive mathematics

The next big advance in robotics will not come from hardware, but from mathematics. Today, robots react to inputs and adapt in real time. Tomorrow they will think ahead.

New mathematical methods such as dual numbers and so-called jets - models for the simultaneous description of movements and their derivatives - are currently quietly but fundamentally changing our understanding of how change is recorded mathematically. They enable systems to calculate not only what happens during a robot movement, but also how this movement affects the dynamics, forces and subsequent states in the overall system. This leads to faster optimization, more comprehensive scenario planning and adaptive control that is almost intuitive.

Robots that calculate the effects of a path correction before executing it or simulate several "what-if" scenarios within milliseconds are conceivable. This is not science fiction, but a consistent further development of modern derivation and forecasting methods.

In my view, predictive intelligence will define the next generation of automation. The question is not whether this change will come, but when and who will lead it.

2. from standalone to line: networked cobots

Imitation learning - i.e. learning by observing human or robotic role models - will become a key skill of the next wave of automation. Today, most robots work as individual instances, controlled by central fleet systems or rigid programs. In the future, they will learn from each other and from humans, partly guided, partly autonomously, and join together to form adaptive teams that share experiences, movement patterns and strategies in real time.

This development is based on research approaches in which robots not only follow a predefined trajectory, but also observe, imitate and jointly refine actions. This creates dynamic coordination without rigid scripts. Although industrial robotics providers have already laid important foundations with fleet management and synchronized multi-arm systems, true peer-to-peer learning and self-organization are still in their infancy. Nevertheless, I am convinced that by 2026 we will see the first real applications based on imitation-learned models of physical AI in which perception, movement and decision-making are closely interlinked.

The advantages are obvious:

• Faster configuration and reconfiguration of processes by learning from demonstrations instead of classic programming
• Greater robustness in the event of unexpected changes, such as varying components, tolerances or process sequences
• More natural human-robot collaboration, in which robots interpret human intentions or the rhythm of a leading robot and follow them adaptively

With the further development of safety standards, inter-robot communication and orchestration tools, this form of collaboration is becoming industrially scalable. As a result, robots are evolving from isolated units into cooperative, continuously learning teams.

3. dedicated AI instead of general-purpose solutions

Anders Billesø Beck is Vice President AI Robotics Products at Universal Robots. © Universal Robots

Instead of generic AI platforms, manufacturers will increasingly rely on task-specific AI applications - i.e. solutions that have been developed specifically for individual processes such as welding, grinding, inspection or assembly. AI welding, AI finishing, AI assembly and AI inspection will become standard components of new robot cells. Automation is thus finding its way into processes that were previously considered too variable or too complex.

These vertical applications are pre-trained, pre-integrated and can be used productively right from the start. A prime example is welding, where AI-supported seam tracking using image processing and machine learning for automatic parameter optimization is already improving quality, stability and reproducibility.

The next frontier lies in complex, fine-motor tasks such as assembly, screwing and sophisticated handling - areas that have traditionally been difficult to automate. In industrial applications, AI will enable robots to better deal with component and process variability. In service industries, similar approaches are increasingly being used for packaging, sorting or sensitive material handling.

Significant progress has also been made in logistics in recent years: AI-supported robotic systems are now taking over picking, accumulation and touch processes efficiently and scalably. By 2026, investments are expected to increasingly shift from the logistics sector to the retail sector - the next step in bringing robotic automation closer to everyday life.

4. data-driven value creation

The next big change is not just about the movement or intelligence of robots, but the way in which their data creates economic and technological value. Today, most of the extensive data - sensor data, image information or force profiles - remains directly on site at the customer's premises. This is ideal for data protection and reaction speed, but significantly limits access to real training data for AI developers.

In the future, robot manufacturers will increasingly become operators of secure, voluntary data exchange platforms at manufacturer or ecosystem level. With customer consent and subject to strict data protection requirements, anonymized performance data could be aggregated and made available as training data sets or model-based services. For example, welding robots that share anonymized quality data from seams or grinding robots that contribute surface parameters are conceivable as the basis for more intelligent AI for automated fault detection, predictive maintenance and adaptive process control.

The real opportunity lies in transforming raw telemetry data into structured, privacy-compliant insights that accelerate innovation across the ecosystem. For manufacturers, this means new revenue streams and continuous improvement of their own systems. For users, this means more powerful AI tools, trained under real-life conditions, without revealing sensitive information.

The result is a self-reinforcing cycle of innovation: every robot used contributes to making the next generation more intelligent.

The return on predictive robotics in industrial applications

The future of robotics will be shaped by the interplay of advanced methods, intelligent applications and data-driven strategies. Modern mathematical processes give robots the ability to think ahead and adapt dynamically. Leader-follower concepts transform previously isolated machines into cooperative teams that flexibly reconfigure workflows. Vertical AI applications provide ready-to-use intelligence for specific tasks and reduce rework while improving quality. This is complemented by a new data economy in which anonymized field data continuously enables better AI models.

Together, these developments lead to a significant leap in mission ROI - i.e. the economic benefit per robot application: higher productivity per robot hour, faster commissioning and retooling, less downtime and continuous improvement based on real application data.