'Humanoid robots are the new frontier for Italian industry'

The new frontier of artificial intelligence is physical. It is about creating 'embodied' models that can read the environment around them to make decisions and take action. Technologies designed to be exploited by all industrial robotics chains. They will obviously be essential for self-driving cars, real robots on wheels. And they will be essential for all automata, whether specialised or generalist, humanoid or inspired by other forms. For Italy, this is a decisive frontier. How will Italian industrial robotics tackle it? How will Italy's remaining car manufacturers explore it? And how will small and medium-sized Italian manufacturers interpret it?

The Istituto Italiano di Intelligenza Artificiale per l'Industria (AI4I), the Turin-based foundation set up to accelerate the growth of national expertise in the industrial applications of artificial intelligence, is dealing with this in depth. Yesterday at the OGR in Turin, this was the topic of the first public seminar of the CSP IAS, the Institute for Advanced Study created within AI4I thanks to the Compagnia di San Paolo Foundation. Speakers included Marco Pavone, new chairman of the AI4I Scientific Committee, associate professor of Aeronautics and Astronautics at Stanford University and Senior Director Autonomous Vehicle Research at NVIDIA, and Riccardo Mariani, Vice President, Safety, at NVIDIA.

Pavone is pragmatic. In the world of physical intelligence, the United States and China are ahead, but for Europe, and Italy, there are important opportunities. But we need to understand what is happening. How does European industrial robotics fit into this context? "Industrial robots are a well-known business, worth many billions. But humanoid robots are a potential business worth trillions,' explains Pavone. 'In this context, spaces are important to build infrastructure and very profitable applications and there are very good opportunities for Italians as well.'

Of course it is about change. "One of the strengths of industrial robotics manufacturers lies in the fact that over many years of experience they have accumulated an exceptional amount of knowledge and data, which could prove essential for the development of new models. What about the sports cars that are produced in Italy? 'I know those companies well. They will probably want to introduce semi-automatic systems, but not replace the driver's experience'. These are driving assistance systems, which increase safety, but do not allow the car to be completely in charge of driving. "The business of fully autonomous cars is rather developing in the form of new public transport services, as in the case of driverless taxis in San Francisco, Los Angeles and Phoenix". Developments consistent with the strategy of those who conceive of artificial intelligence as an infrastructural technology that Pavone sees as a winner. The greatest opportunities are to be found in this context: probably offering data sources, specialised mechanical skills, design, and application ideas to be imagined.

In this context, large language models (llm) are the forms of artificial intelligence that have undergone the greatest development in recent years. And they can also serve robotics. "The applications of llms in this field range from writing software for robot on-board systems to providing sophisticated reasoning systems to deal with complex situations. They can even become useful as parallel control systems, to monitor the efficiency of autopilots and recognise anomalies'.