Innovation leverages physical AI, a major technological breakthrough with potential benefits for the sector’s performance.
Orano, a recognised industrial leader in the recovery and transformation of nuclear materials and Capgemini, an AI-powered global business and technology transformation partner, have announced the deployment of the first intelligent humanoid robot in the nuclear sector. This project marks a major step forward for a strategic industry that has long been a pioneer in innovation.
Deployed at the Orano Melox Ecole des Métiers in the Gard region of France, the robot named Hoxo is equipped with embedded Artificial Intelligence and advanced sensors for real-time perception, autonomous navigation, execution of technical gestures and interaction. Its purpose is to replicate human movements and operate alongside teams within nuclear facilities, including in challenging intervention environments.
Over the next four months, Orano Melox’s innovation teams will conduct a testing phase to validate the robot’s range of applications, combining mobility, precision and Artificial Intelligence. By offering an agile, scalable robotic platform, this initiative is expected to enhance industrial performance and potentially support operators through robotic assistance.
“Hoxo opens new perspectives for our operations by combining an intelligent and ergonomic robotic solution with the expertise of our on-site teams. It’s an innovation we aim to evolve to meet our industrial needs, contributing to both safety and competitiveness as we tackle the challenges of today and tomorrow,” said Arnaud Capdepon, Director, Orano Melox.
“This project, led by our AI Robotics and Experiences Lab, embodies the convergence of robotics, Artificial Intelligence, computer vision and digital twins. It redefines human-machine interaction in sensitive environments and pushes the boundaries of industrial automation. Through this initiative, we harness the potential of physical AI to address Orano’s most demanding industrial challenges,” said Pascal Brier, Chief Innovation Officer, Capgemini and member of the Group Executive Committee.
Embodying intelligence: The challenge of Physical AI in the real world
Cambridge Consultants, the deep tech powerhouse of Capgemini, are currently exploring the next step in the potential role of humanoid robotics and physical AI.
Ali Shafti, Head of Human-Machine Understanding, Cambridge Consultants, said Artificial Intelligence is stepping out of its digital confinement and gaining the senses and limbs necessary to understand and effect change in the real world.
“This transition has sparked a significant debate on the ideal form factor for Physical AI across industrial environments.
“On one side, special-purpose robotics offer targeted efficiency for specific environments and narrowly defined tasks. On the other, general-purpose robots, such as humanoids, promise the flexibility to operate in diverse, changing spaces that have been designed primarily for people rather than machines.
“The reality is that there is likely no single ‘winner’. Some use cases do not call for the complexity of a general-purpose solution and are better served by specialised systems that are optimised for repeatability and constrained environments.
“However, at Cambridge Consultants, researchers have chosen to run their research and development efforts in the space of Physical AI with a focus on humanoids precisely because of the immense challenge they present.
“Humanoids are significantly harder to control than standard robotic platforms. They must balance on two legs while manipulating payloads, relying on complex whole-body dynamics to remain stable. These challenges are compounded when operating in real-world industrial settings that are unpredictable and often hazardous.
“By deliberately tackling these ‘hard problems’, from robust locomotion to precision handling, engineers drive advancements that benefit the entire spectrum of Physical AI, regardless of the eventual form factor. Progress made in humanoid research often cascades into improvements across other robotic systems.
“The work focuses on three distinct but interconnected streams: whole-body dynamics, fine manipulation and human-robot interaction. Each stream addresses a critical barrier to deploying intelligent machines in environments where humans and robots must coexist safely.
“In whole-body dynamics, researchers are bridging the ‘sim-to-real’ gap. Using supercomputers to run thousands of simulated environments simultaneously, they employ a combination of reinforcement and imitation learning to train 23 degrees-of-freedom to move in unison. This approach allows motion policies to be deployed that remain stable in the chaos of the physical world.
“Simultaneously, fine manipulation teams are addressing the scarcity of high-quality training data. By fine-tuning foundation models such as NVIDIA’s Groot and utilising custom tele-operation pipelines, robots are being taught to handle objects with the dexterity required for tasks ranging from parcel sorting to shelf restocking.
“However, physical capability is only half the equation. Whether a robot has two legs, four wheels or a stationary arm, it must possess social intelligence to work effectively alongside people in shared environments.
“As explored in work on Physical AI and human-robot interaction, effective collaboration requires a shared context. Systems are moving beyond basic command-and-control interfaces to environments where users can instruct robots through natural mixes of verbal and non-verbal cues.
“This is where Physical AI meets Social AI. It is the ability to ‘read the room’, not just the map, and to respond appropriately to subtle human signals such as posture, gesture and tone.
“Understanding must be reciprocal. Humans naturally adapt to one another and robots must do the same. This necessitates Human-Machine Understanding, structured models of human behaviour in the context of the environment and the task at hand.
“As outlined in analysis of why adopting Human-Machine Understanding positions businesses for the future, this approach bridges the gap between rigid world models and fluid human cognition. It enables machines to anticipate needs rather than merely react to commands.
“By mastering the complexities of humanoids, from maintaining balance to maintaining trust, researchers are building the foundational capabilities for the next generation of intelligent machines, whatever form they may take. These developments are expected to play a key role in accelerating Digital Transformation across industrial sectors, including energy, manufacturing and infrastructure.”
