Artificial touch and energy efficiency: a new frontier for AI

Artificial touch and energy efficiency: a new frontier for AI

A study coordinated by the Scuola Superiore Sant'Anna in Pisa proposes a bio-inspired tactile system for robots and intelligent machines with low energy consumption.. This is a significant breakthrough for research into artificial touch, bringing to the forefront the challenge of energy efficiency and sustainability in the future of artificial intelligence

The future of artificial intelligence will depend not only on its intelligence and data processing capabilities, but also on its energy sustainability. A study published in Nature Communicationsand coordinated by Scuola Superiore Sant'Anna, in collaboration with STMicroelectronics, University of Zurichand ETH Zurich, has developed an innovative bio-inspired tactile system that reproduces the perception and processing of human touch with drastically reduced energy consumption.

This represents a significant breakthrough in artificial touch research, enabling the development of intelligent robots and machines capable of interacting and collaborating effectively with the surrounding environment, and prompting a radical rethinking of the future of artificial intelligence placing the challenge of energy efficiency and sustainability at its core.

A tactile system inspired by human skin and the human nervous system

The system integrates a sensorised artificial skin, equipped with optical sensors that mimic the behaviour of the tactile receptors of human skin, with a bio-inspired neural architecture. The responses generated by these sensors are then processed by a neural network that reproduces the organisation and strategies of the human nervous system thanks to a special processor, based on hybrid analogue-digital circuits.

The solution proposed in the study allows tactile signals to be processed not only more accurately than conventional artificial intelligence approaches, but also considerably more energy-efficiently. This bio-inspired system paves the way for a new generation of effective, low-power sensory systems for the robotics of the future.

New perspectives for robots, neuroprostheses and human-machine interaction: towards a new paradigm for AI

The study is expected to have a significant impact across multiple application domains: firstly, in a social context where robots are expected to interact increasingly with humans, the technology could be used in the development of bionic neuroprostheses, collaborative robots and humanoids.

The possibility of distributing tactile sensors over large surfaces while maintaining extremely low energy consumption represents a decisive step towards the development of truly effective and sustainable robots, even in implantable, wearable and mobile applications.

Furthermore, the study offers a concrete response to one of the most pressing challenges facing artificial intelligence: energy sustainability. Today, artificial intelligence systems, such as large language models and generative and agentic AI systems, require increasing amounts of energy, with rapidly rising economic and environmental costs. Forecasts indicate that this trend is set to intensify, making the current technological trajectory increasingly unsustainable if it continues to rely exclusively on ever-growing supercomputing power. The principles demonstrated in this study point to a possible paradigm shift, based on the need to radically rethink the way intelligent systems are built, drawing inspiration not only from the capabilities of the human brain and body, but above all from their extraordinary efficiency.

Statements

Andrea Ortone, first author of the study, PhD student at the Scuola Superiore Sant'Anna: By replicating the spike language of the nervous system, we have achieved accurate and efficient artificial tactile sensitivity based on analogue circuits that emulate biological neurons to decode external stimuli with drastically lower energy consumption than conventional artificial intelligence. A central challenge was translating the distributed and dynamical nature of touch sensing into an efficient artificial system, achieved through the tight integration of sensing, neural modelling and neuromorphic hardware within a unified spiking architecture.

Calogero Oddo, study coordinator and professor in charge of the Neuro-Robotic Touch Laboratory at the Scuola Superiore Sant'Anna: The sense of touch is particularly complex because it is the result of physical interaction, distributed and variable over time, between the human body and the surrounding environment, with a multitude of fundamental scientific questions still to be clarified. This study demonstrates the potential applications of neuroscientific research on the sense of touch, including in collaboration between universities and a highly innovative company such as STMicroelectronics, with an impact on the development of neurotechnologies for bionics, medical, service and industrial robotics. The neuro-inspired artificial intelligence architectures of this research will be integrated into robots and machines in the future.

Giuseppe Desoli, co-author of the study, Company Fellow, AI HW Architectures R&D Director STMicroelectronics: This work describes a revolutionary technology that is set to play a key role in the future integration of ultra-low-power sensors with advanced intelligence, enabling sensors to provide semantically relevant information rather than raw data, using algorithms and architectures inspired by biology. These advances open up new scenarios in which sensors can intelligently filter information directly after acquiring analogue signals, operating with extremely low power consumption. This capability supports long battery life and the use of energy scavenging systems, as well as a significant reduction in the amount of data to be transmitted. STMicroelectronics remains committed to promoting open innovation through close collaboration with academia and research institutes. This work exemplifies the synergies possible when research results can be transferred into real industrial products.

Giacomo Indiveri, co-author of the study, Institute of Neuroinformatics, UZH e ETHZ: A crucial aspect of this work lies in the biophysically realistic "NeuroAI" approach adopted to develop a biologically plausible (i.e. "biomimetic") neural network model, initially simulated on a computer. The fact that this model was also validated using a neuromorphic processor with hybrid analogue-digital circuits further highlights its robustness and highly innovative nature. This work demonstrates how the approach taken can lead to robust real-time processing based on analogue neuromorphic circuits which, like their biological counterparts, are noisy and highly variable on the one hand, but extremely energy-efficient on the other.

Gallery