Interview With Zhaopeng Chen, CEO of Agile Robots Physical AI is Reolutionizing Industrial Robotics

Related Vendor

SIMTOS (Seoul International Machine Tool Show)

What must a humanoid robot be able to do? What are the biggest challenges? What role does an appropriate ecosystem play? Zhaopeng Chen, co-founder and CEO of Agile Robots, answers these questions and many more in an interview with at—aktuelle technik.

at-aktuelle technik: Agile Robots is one of the few European unicorns in the field of robotics. As a co-founder, what is your vision for robotics?

Zhaopeng Chen: Our vision at Agile Robots is to make robots not only autonomous, but also adaptive, adaptable and highly intelligent. With Physical AI, we want to develop robots that can react just as flexibly as humans in the physical world—be it in manufacturing, logistics or complex service applications. We envision a future in which robots are true partners for humans, responding dynamically to change and continuously improving.

At the moment, all media channels seem to be flooded with dancing and kung-fu fighting humanoid robots, especially from China. Agile Robots focuses more on industry. What do you think a humanoid robot should be able to do?

Z. Chen: Above all, a humanoid robot must be able to react flexibly to changing environments. Our specially developed Robotics Foundation models are crucial for this: they form the 'brain' of the robot, enable continuous learning from the environment, coordinate movements and actions in real time and ensure that the robot can adapt independently to new tasks and scenarios. The hands also play a central role: our humanoid robot Agile One has particularly dexterous hands with 21 degrees of freedom, which enable precise fine motor skills and thus cover a wide range of industrial applications. However, it is important to note that humanoid robots are only part of the solution. At Agile Robots, we see them as one element of a comprehensive ecosystem in which all systems are intelligently networked to create real added value in industrial processes. Our focus is on ensuring that robots not only look impressive, but can also be used practically, safely and adaptively in real working environments.

In which tasks do you see the greatest potential?

Z. Chen: We see particularly great potential for humanoid robots in sectors with highly standardized or recurring work processes, such as the automotive industry, consumer electronics, logistics and mechanical and plant engineering. Humanoid robots are by no means intended to replace existing automation solutions: In many areas, robotic arms or autonomous mobile robots will continue to be the more efficient choice. Humanoid robots can be used, for example, where flexibility, adaptability and interaction with complex environments are required—especially in places that are only easily accessible to humans due to narrow passages, steps, doors or complicated construction methods. Here, they can take on tasks that are difficult for traditional automation solutions, relieve employees and make processes more efficient while increasing flexibility in production.

Many developments in the field of humanoids are still not fully developed. What are the biggest challenges here?

Z. Chen: One of the biggest technical challenges in humanoid robotics is currently the availability and quality of data. Although our robotics foundation models already enable advanced perception and decision-making capabilities, they are heavily dependent on extensive, diverse data sets in order to work reliably in real production environments. Unlike language models, which can be trained with large amounts of freely available Internet text, robotic systems require multimodal, physically anchored data—such as image, force and movement information. These are more complex to capture and standardize and to provide in sufficient quantities. In addition, many industrial data sets are strongly tailored to specific tasks or environments. They are often not diverse enough to be able to reliably transfer models to different environments and processes or product variants. We are therefore investing heavily in scaling our data basis: by combining real production data from our data farms and factory environments with synthetic simulations and human demonstrations—with the aim of training robust and scalable models for a wide range of industrial applications.

What are the technological features and solutions that distinguish Agile Robots' humanoid robots from those of the competition?

Z. Chen: The Agile One humanoid robot uses our own Robotics Foundation models, which have been trained with real data from factories and data farms, synthetic simulations and human demonstrations. In contrast to generic data sets, which are usually very limited or tailored to standard tasks, Agile One benefits from a particularly extensive and diverse data basis that ensures its precision, adaptability and performance in a wide variety of industrial environments. Thanks to this basis, the robot can grip objects safely and autonomously—supported by its particularly dexterous hands with 21 degrees of freedom, which reliably enable even complex manipulations. In addition, Agile Robots is not only the manufacturer of Agile One, but also offers a comprehensive portfolio of robotics solutions. As a result, customers benefit from an integrated ecosystem that enables them to select the optimum solution for their individual application. The Agile One humanoid robot can work together with other robotics solutions to master complex production tasks efficiently and flexibly.

Agile Robots not only builds humanoid robots, but also intelligent collaborative robots and robotic systems specialized for specific industrial applications. How important is such a basis?

Z. Chen: This is exactly what sets Agile Robots apart: Thanks to our broad portfolio, we can tailor solutions precisely to individual requirements and ensure that Agile One and other systems work together seamlessly. We are convinced that this integrated ecosystem is crucial to fully realize the benefits of our robotics technology. It enables not only the optimal use of Agile one, but also coordinated collaboration with other robotics solutions. This allows processes to be scalable, flexible and robust—an advantage that isolated systems, such as a single humanoid robot, cannot offer.

How do you see the interaction between hardware and software? What is more important?

Z. Chen: Hardware forms the basis of every robot—precision, stability and maneuverability are crucial. But without the right software, robots remain static machines. Only through intelligent control, flexible algorithms and seamless integration do they become powerful, adaptive solutions. At Agile Robots, we consistently view hardware and software as a single unit. Our Agilecore software platform is at the heart of this approach. It not only controls individual robots, but also orchestrates the collaboration of multiple systems within an integrated ecosystem. This allows processes to be coordinated intelligently, adjustments to be made in real time and new tasks to be implemented quickly—all via a central, intuitive platform. In this way, we transform robots from isolated machines into networked, flexible production and service units that create real added value.

How has AI integration into robots developed in recent years?

Z. Chen: In recent years, the integration of artificial intelligence into robotics has developed considerably—moving away from rigid, programmed processes towards adaptive, self-learning systems. At Agile Robots, we rely on physical AI. Our robots no longer rely solely on predefined motion sequences, but are able to perceive, interpret and react to their environment in real time. This means that they can adapt to changing conditions, optimize processes and make decisions based on actual circumstances. Physical AI thus transforms robots from simple automated tools into adaptive systems that work effectively even in dynamic and unpredictable environments.

What role does machine learning play in robot autonomy?

Z. Chen: Instead of simply executing rigid instructions, autonomous robots are now learning to perceive, interpret and react flexibly to their environment with the help of AI. Sensor and multimodal data from cameras, touch sensors and other sources are processed using deep learning models so that robots can recognize patterns, make decisions and perform precise actions—similar to how a human nervous system translates sensory impressions into behaviour. Specifically, techniques such as neural networks and machine learning approaches—including reinforcement learning and imitation learning—enable robots not only to 'execute' tasks, but also to learn from experience, develop optimal strategies and act reliably even in dynamic, unstructured environments. In conjunction with our Robotics Foundation models, they form the cognitive control system that seamlessly links perception, contextual understanding and autonomous action of a robot.

How reliable are the robots in safety-critical applications? What needs to be considered here?

Z. Chen: Safety is the top priority for our robots, especially in applications where people work in close proximity. Our robots constantly monitor their speed and movements and strictly adhere to defined limit values. They move exclusively within defined work areas and limit both force and kinetic energy to minimize the risk of unforeseen contact. In the event of a collision, the robots react within milliseconds, immediately stop all movements and switch to a safe state. In addition, easily accessible emergency stop switches are integrated so that all actions can be interrupted immediately if necessary. Furthermore, the robots use spatial awareness to continuously monitor their surroundings. They recognize obstacles, people and other objects and proactively adjust their movements and forces to avoid collisions. This combination of mechanical protection mechanisms, intelligent monitoring and precise environmental detection ensures that our robots operate reliably, predictably and in a human-friendly manner, even in safety-critical environments.

Before founding Agile Robots, you developed your first humanoid hand at DLR. What was the biggest challenge you faced?

Z. Chen: One of the key tasks was to find the right balance between agility, precision and robustness. The aim was to develop a hand that could not only perform sensitive, human-like movements, but was also stable and powerful enough to securely grip and manipulate a wide variety of objects. At the same time, the actuators could not be too large or heavy so as not to impair the natural movement behavior and maneuverability of the design. Mechanics, sensors and control technology had to be coordinated during development in such a way that the hand combines human-like movements with a high degree of freedom and sensitive control. We integrated sensors for measuring torque and position into each finger joint to enable real-time feedback and active compliance control. At the same time, we focused on a modular and easy-to-maintain design to facilitate customization, configuration and manufacturing. This combination of mechanical complexity, precise control, intelligent sensor technology and robust construction made the development of the hand particularly challenging - and at the same time extremely exciting.

Today, Agile Robots has the Agile Hand. What has happened since the initial development?

Z. Chen: Since its initial development, the Agile Hand has been continuously refined. There have been many iterations, the hand has become more compact and the kinematics of the fingers and thumb have been significantly improved to enable even more precise and natural movements. One particularly important advance is the use of AI-supported control: the hand can now train and adapt movements and gripping techniques independently. It is therefore no longer limited to predefined sequences, but can react flexibly to new objects and situations—a decisive step towards intelligent, human-like robotics.

How do you view the scalability of robots and what growth strategies does Agile Robots have?

Z. Chen: We automate entire factories—scalability plays a key role here. We offer the full range of solutions in terms of both hardware and software. This means that different robot types and automation systems can be flexibly combined depending on the application and used in a wide variety of production environments. Physical AI also makes robots more intelligent—they are more aware of their surroundings, learn from data and can perform complex tasks increasingly autonomously. For Agile Robots, this is a decisive step towards making automation even more flexible and usable on a large scale. As a company, we are committed to the continuous development of our technologies with a focus on physical AI, international expansion into key industrial markets and the expansion of our ecosystem through partnerships and strategic acquisitions.

In which application areas do you see the greatest potential for growth?

Z. Chen: We see particularly great potential for robots in industries with highly standardized or recurring work processes, such as the automotive industry, consumer electronics, logistics and mechanical and plant engineering. In these sectors, robots can not only increase efficiency, but also improve the quality and precision of products. By using automated systems, production processes can be reliably standardized, error rates reduced and bottlenecks avoided. At the same time, robots enable greater flexibility within existing processes: they can be quickly adapted to new products or changed production volumes without the need for extensive retooling.

How will robotics develop in industry in the future and what role will humans play?

Z. Chen: In the future, industrial robotics will be characterized above all by "physical AI"—in other words, by the combination of artificial intelligence, sensor technology and robotics. This will enable robots to perceive and understand their environment and react to it independently instead of just executing pre-programmed sequences. This makes them more flexible, more autonomous and able to take on complex tasks in production. At the same time, increasingly networked production systems are emerging in which robots, software and AI work together to optimize entire production lines. The greatest added value lies not in the individual robot, but in the intelligent interaction of all systems within the factory. Humans will not be replaced. Instead, their role is changing. While robots take over monotonous, physically demanding or risky tasks, the role of humans is shifting more towards monitoring, improving and further developing production processes. In this way, intelligent automation can also help to keep industrial locations such as Germany competitive in the long term.

Design simulation for industrial robots

How to make your production more flexible with robotics