When Machines Understand What They're Doing Five Technologies Behind Physical AI

Related Vendors

FAULHABER Drive Systems

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

Physical AI is considered the next step in the evolution of industrial digitalization, yet the term often remains abstract. What exactly enables machines to understand their environment and act independently? It is the interplay of a few key building blocks that actually transforms data-driven systems into capable agents. NTT Data reveals the five technologies required to achieve this.

For a long time, artificial intelligence was limited to the analysis of data. With Physical AI, this approach is expanded to include a new dimension: the systems directly sense their environment using cameras, lidar technology, and other sensors. This enables them to interpret the real world and respond flexibly to changes. Unlike traditional automation, it is no longer necessary to explicitly define or program all rules and scenarios in advance. However, for this form of AI to work, it is not enough to simply further develop existing technologies. Physical AI emerges only where multiple technological disciplines intertwine to form a closed-loop system of perception, interpretation, and action. Five building blocks are particularly crucial in this context:

Robotics in Automotive Manufacturing

Humanoid Robot Will Sort Components at BMW in the Future

• 1. Sensors and sensor fusion as the foundation of perception:Physical AI begins with the ability to accurately perceive the real world. Modern sensor systems take on the role of a finely tuned nervous system: They measure temperature, pressure, vibrations, spatial positions, or optical properties and convert physical states into digital signals. However, it is not the individual sensor that is crucial, but rather the interaction of many data sources. Only through sensor fusion does a consistent overall picture emerge, allowing conclusions to be drawn about complex relationships. In industrial practice, this means, for example, that the states of machines can no longer be evaluated in isolation but must be interpreted in the context of multiple parameters. This capability is essential for reliably detecting anomalies and enabling adaptive processes.
• 2. Edge Intelligence and Energy-Efficient Chips for Real-Time Decisions:Processing large amounts of data directly where it is generated is a key prerequisite for Physical AI. Modern sensors and embedded systems increasingly have their own computing capabilities and run machine learning models locally. This shift of intelligence to the edge of the network significantly reduces latency and enables responses in milliseconds. This requires highly specialized, energy-efficient chips capable of running even complex AI models with limited resources. Such architectures are optimized to immediately recognize patterns in data streams—for example, to identify wear and tear early on or to correct quality deviations directly within the production process. At the same time, the need to centrally transmit large amounts of data is reduced, which saves both bandwidth and energy.
• 3. Artificial Intelligence as the Analytical Core:The true value created by Physical AI stems from its ability not only to collect data but also to transform it into actionable knowledge. This is where machine learning, pattern recognition, and—increasingly—generative AI models come into play. In particular, foundation models and so-called vision-language-action (VLA) models are fundamentally transforming the development of intelligent environments. Instead of training each application from scratch with large amounts of specific data, companies can build on pre-trained models and adapt them specifically for complex use cases. This significantly shortens development times and allows new use cases to be implemented more quickly. VLA models, in turn, interpret visual information, language, and context together. This means that machines can understand spoken commands, react to unexpected situations, and trigger specific actions for robotic arms or actuators. However, traditional AI methods remain indispensable for deterministic control, regulation, and safety tasks. The greatest potential arises where both approaches work together: generative models handle perception, contextual understanding, and planning, while traditional AI methods ensure precise and reliable execution.
• 4. Connectivity via 5G/6G and industrial communication standards:For physical-AI systems to operate consistently, data must be transmitted securely at high speeds and with minimal latency. Modern communication infrastructures such as 5G and, in the future, 6G lay the foundation for this. They enable reliable, low-latency networking of machines, sensors, and IT systems—even in highly dynamic production environments. At the same time, standardized protocols and data models are gaining importance, for example, for the semantic description of machine data and to ensure interoperability. Only when different systems speak the same “language” can data be efficiently integrated and processed further. This end-to-end connectivity is crucial for avoiding isolated silos and creating a holistic view of production processes.
• 5. Advanced Robotics and Adaptive Actuators:While sensors and AI handle perception and decision-making, physical execution is handled by modern robotic systems. New generations of robots are no longer limited to rigid sequences but react flexibly to their environment. This is made possible by improved actuators, more precise control systems, and the close integration of AI models. Collaborative robots, autonomous transport systems, and adaptive grasping systems are capable of dynamically adapting their movements and operating reliably even in unstructured environments. This completes the control loop: systems sense their environment, interpret the data, and immediately translate decisions into physical actions.

Interview With Zhaopeng Chen, CEO of Agile Robots

Physical AI is Revolutionizing Industrial Robotics

“In traditional manufacturing environments, the mechanical interaction of machines was the decisive factor for efficiency and productivity for decades. Production volumes, cycle times, and material flow determined what happened on the shop floor. While the perspective hasn’t fundamentally shifted today—because results still matter— However, our overall understanding of automation has changed,” explains Oliver Köth, Managing Director of Technology & Innovation for the DACH region at NTT Data. “With Physical AI, we can create systems that independently handle uncertainty and variance and use that information to make the right decisions. Such capabilities are particularly in demand in the manufacturing sector. This requires the interplay of several technological developments. Only when sensor technology, AI, connectivity, and robotics work together seamlessly can systems emerge that understand their environment and act autonomously.”

About NTT Data Group

NTT Data is part of the NTT Data Group in the DACH region. With annual global revenue of over $30 billion, it is a leading provider of business and technology services, with 75 percent of the Fortune Global 100 among its clients. Across the group, NTT Data is committed to its clients’ success and strives to positively transform society through responsible innovation. It is one of the world’s leading providers of AI and digital infrastructure. It also offers unique capabilities in the enterprise-scale deployment of AI, cloud, security, connectivity, data centers, and application services. Through consulting and industry-specific solutions, NTT Data supports businesses and society in safely and sustainably moving toward a digital future. As a Global Top Employer, the group has experts in more than 70 countries. It also offers customers a robust ecosystem featuring innovation centers as well as established partners and startups. NTT Data is part of the NTT Group, which invests more than $3 billion in research and development each year.