The New Frontier of Industrial Robotics: Data, Open Platforms and Physical AI

Robots are no longer isolated machines and are becoming systems capable of perceiving, learning, deciding and acting on the physical world.

Over the past few years, artificial intelligence has advanced mainly within the digital world: interpreting texts, images, conversations and large volumes of data. The next step is to bring this capability into the physical world.

The combination of artificial intelligence, robots, sensors, autonomous vehicles and industrial systems is giving rise to what is known as Physical AI, or physical artificial intelligence: systems capable of perceiving their environment, making decisions and executing actions on real-world objects, processes and spaces.

The sessions by Botiful.ai, PAL Robotics and LAAS, Agibot and KUKA at Viva Technology allowed us to identify four profound changes: value is shifting from hardware to software and data; robots are learning to interact safely with people; open platforms are replacing automation islands; and natural language is beginning to become a new interface for controlling physical systems.

For INDPULS’s industrial partners, it is not only about anticipating when humanoid robots will arrive. It is about understanding what infrastructure, use cases and capabilities need to start being built so that the new generation of robotics can generate real value.

1. Value is shifting from the robot’s body to its brain

Botiful.ai starts from the idea that the hardware of humanoid robots will eventually become standardized and be produced at scale by major manufacturers, particularly Asian ones. For this reason, the company is not focusing its efforts on manufacturing the robot, but on developing its “brain”.

This brain includes conversational artificial intelligence, language understanding, decision-making, integration with business systems and autonomous navigation. The goal is for the robot not simply to talk, but to understand a request, consult corporate information and execute an action.

In a demonstration applied to a hotel reception, the robot could welcome the guest, locate the reservation, verify their identity using an official document and facial recognition, assign a room and send the access code to their mobile phone. The entire process was carried out through natural interaction.

The most relevant aspect is not the humanoid appearance. It is the integration with applications the company already uses, such as Salesforce or other management platforms. The robot stops being a conversational assistant and becomes a new interface for accessing business processes.

Botiful.ai is also developing localization and mapping capabilities —SLAM— so that robots can move autonomously through indoor spaces, avoid obstacles and interact safely with people.

This approach leads to a first conclusion: mechanical capabilities will remain important, but the differentiating value will increasingly lie in the robot’s ability to understand context, connect to data and systems, execute processes and learn.

2. From rigid automation to safe physical collaboration

Traditional industrial robotics has mainly been designed to repeat precise movements within spaces separated from people. The new generation of robots seeks to share the environment with operators and adapt to physical contact.

PAL Robotics and LAAS work with systems based on torque control, a technology that makes it possible to continuously regulate the force applied to an object or a person. Instead of simply following rigid trajectories, the robot can react to contact and adjust its movement in real time.

This capability makes it possible to tackle operations requiring sensitivity, precision and adaptability. The technology is already being applied in demanding industrial processes, such as working on Airbus aircraft components.

The demonstration of the TALOS humanoid robot, developed by PAL Robotics, shows this evolution. The robot is designed for industrial applications and can move autonomously in environments shared with people, with safety as a central development criterion.

The advances are not only algorithmic. New mechanical actuation systems are also being developed to support higher loads and reduce motor overheating problems that limit many current robots.

At the same time, reinforcement learning is accelerating locomotion. According to the speakers, tasks that could require several years of development using conventional techniques can be achieved in approximately one year through machine learning.

3. Open standards accelerate the ecosystem

PAL Robotics, LAAS and KUKA share a commitment to ROS2, one of the leading open standards in modern robotics. Working on a common foundation makes it easier for researchers, companies and developers to build new applications without having to create the entire infrastructure from scratch.

Interoperability reduces development times and prevents each manufacturer from building a completely closed ecosystem. It also makes it possible to incorporate new sensors, artificial intelligence models, simulation tools or applications developed by third parties more quickly.

In Europe, this philosophy forms part of the ambition to build its own robotics ecosystem. Projects such as HyQReal, AgileX and T-REX, within the framework of France 2030, connect universities, research centres and companies to compete with major North American and Asian players.

The speed of change makes it increasingly unrealistic for a single company to develop all the required capabilities. The future points towards open architectures in which manufacturers, integrators, research centres and user companies share components and knowledge.

4. Data could be the main competitive advantage

Agibot offers a different perspective on humanoid robotics. The Chinese company, founded in Shanghai, has a broad portfolio: humanoid robots, specialized equipment for factories, industrial cleaning robots and quadrupeds for safety inspections, disaster-zone interventions or material transport in hazardous environments.

These robots combine mobility with cloud-based generative AI models, allowing them to maintain natural conversations. But the company’s most significant approach is its learning model.

Each deployed robot captures information about movements, interactions, errors, environments and tasks. This data continuously feeds the platform’s large artificial intelligence models and improves the behaviour of the entire fleet of robots.

This creates an acceleration loop: more robots generate more data; data makes it possible to train AI better; and better AI makes robots more useful, which in turn encourages further deployment.

According to Agibot, the company was the manufacturer that delivered the most humanoid robots worldwide in 2025. Beyond the stated figure, the strategic message is clear: the ability to scale deployments makes it possible to accumulate real-world data at a speed that is difficult to match.

This helps explain the advantage that the Chinese model can generate. The race will not necessarily be won by whoever happens to manufacture the robot with the best performance, but by whoever is capable of training the best intelligence from the volume and diversity of data from the physical world.

Agibot also recognizes that entering Western markets will depend on collaboration with local integrators and European partners. The technology needs an ecosystem capable of adapting it to the processes, regulations and expectations of each market.

5. From automation islands to a common platform

KUKA starts from a problem present in virtually every factory: automation systems have been added independently over the years. Robots, PLCs, cells and autonomous vehicles use different tools, languages and data. The result is automation islands with limited interoperability and costly integrations.

Its Automation Management Platform —AMP— aims to create a common layer for controlling robots, autonomous mobile vehicles, robotic cells, PLCs and, in the future, humanoid robots, regardless of the manufacturer.

The goal is not to replace each system, but to orchestrate them. The platform is built on ROS2 and is designed as an infrastructure open to robots from other manufacturers and to artificial intelligence agents developed by third parties, including players such as NVIDIA, OpenAI and Google.

This approach points to an important shift: robots may eventually become relatively standardized components, while value becomes concentrated in the software capable of coordinating and managing them and enabling them to learn together.

6. From programming movements to defining intentions

AMP introduces a model based on intent-based operations. Instead of programming each movement in detail, the user defines what they want to achieve: transport a pallet, feed a production line or execute a logistics operation. The platform converts this intention into concrete actions adapted to the available equipment.

This layer can reduce the need for specific programming and make it easier for artificial intelligence to plan physical processes. It also anticipates a new interface for the factory: operators could give instructions in natural language and allow the system to translate them into coordinated actions across different machines.

This does not mean eliminating technical knowledge, but changing the level at which instructions are expressed. Complexity shifts towards the platform, the models and system governance.

7. The digital twin as a learning infrastructure

KUKA’s platform integrates a digital twin of the factory that reflects its status in real time and makes it possible to simulate scenarios before executing them. AI-based analytics, anomaly detection, predictive maintenance and natural-language queries can be incorporated into this virtual model.

The digital twin does not function merely as a visual representation. It becomes a common source of truth on which systems can plan, compare options and learn.

The example of AMR fleet management shows the impact of centralization. Updating the map of a factory with fifty robots could require connecting individually to each piece of equipment using a USB drive and could take weeks. With AMP, the update can be distributed simultaneously across the entire fleet with a single click and completed in less than an hour.

The data generated by the robots feeds the digital twin and AI models. A permanent cycle of data capture, simulation, training and deployment is created: an innovation flywheel that enables processes to improve progressively.

8. What does Physical AI mean for industrial companies?

Physical artificial intelligence cannot be deployed at scale on fragmented industrial infrastructure. Robots will need consistent data, connectivity, interoperable systems and an up-to-date representation of their environment.

Preparing does not necessarily mean starting by buying a humanoid robot. It may mean organizing plant data, connecting assets, reviewing integration standards, advancing digital twins, strengthening cybersecurity or identifying processes that could be automated.

Tasks must also be analysed, not just jobs. The first opportunities will probably be found in repetitive, dangerous, ergonomically demanding operations or operations that are difficult to staff due to labour shortages.

Physical AI will connect sensors, generative AI, collaborative robots, humanoids, autonomous vehicles and control systems through the same data infrastructure. Companies will have to decide how this infrastructure is governed, which systems can make decisions and how human oversight is maintained.

9. A shared agenda for INDPULS

The learnings from Viva Technology open up a concrete line of work for INDPULS. A first opportunity is to create a dedicated space on physical AI and intelligent robotics in which partners identify industrial use cases and share needs.

This work could start with a very practical question: which repetitive, dangerous tasks or tasks affected by staff shortages could be automated with collaborative robots, mobile systems or, later on, humanoid robots?

A second opportunity is to develop an international robotics radar tracking startups, manufacturers and leading centres in Europe, China, South Korea and the United States. The objective would not only be to detect new developments, but to understand which technologies are maturing, which already have real-world deployments and which can generate competitive advantages.

Knowledge sharing between companies is particularly important in this field. Robotics pilots require integration, safety, data and operational changes. Sharing results, difficulties and decision-making criteria can reduce uncertainty and prevent each company from having to follow the same path from scratch.

From the digital world to the physical world

A few years ago, we talked about digital transformation. More recently, generative AI has changed the way we interact with information. The next major step could be integrating this intelligence with machines capable of acting on the real world.

The robotics of the future will not simply consist of adding a new machine to an existing process. It will involve connecting software, data, people and physical systems through an infrastructure capable of learning.

For industry, the challenge is not to wait until the technology is completely solved. It is to start preparing the data, processes, teams and partnerships that will make it possible to adopt it thoughtfully when use cases become mature.

In this transition, collaboration between industrial companies, startups, manufacturers, integrators and research centres will be just as important as the development of the robots themselves.