# Active Digital Twins via Active Inference

**Authors:** Matteo Torzoni, Domenico Maisto, Andrea Manzoni, Francesco Donnarumma, Giovanni Pezzulo, Alberto Corigliano  
**arXiv:** [2506.14453](https://arxiv.org/abs/2506.14453)  
**Date:** 2025-06-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2506.14453v1/

## Key points

- Digital twins are a revolutionary technology for real-time monitoring, simulation, and predictive analysis in engineering and applied sciences.
- Traditional digital twins struggle in uncertain environments due to passive data assimilation.
- The active digital twin paradigm introduces active inference, a neuroscience-inspired Bayesian framework that minimizes free energy by continuously refining generative models through Bayesian updates.
- Active digital twins use proactive decision-making to balance goal-directed utility with uncertainty resolution.
- The key innovation is the ability to dynamically plan actions that minimize expected free energy, enhancing autonomy and resilience compared to passive counterparts.
- A case study of a railway bridge demonstrates how active digital twins can revolutionize asset management by proactively identifying potential issues before they escalate into costly problems.

## Summary

In the realm of engineering and applied sciences, digital twins have emerged as a revolutionary technology for real-time monitoring, simulation, and predictive analysis of physical systems and processes. However, traditional digital twins often struggle in uncertain and dynamic environments due to their reliance on passive data assimilation. To address this limitation, a groundbreaking concept known as the active digital twin paradigm has been introduced in a recent paper titled "Active Digital Twins via Active Inference" by Matteo Torzoni, Domenico Maisto, Andrea Manzoni, Francesco Donnarumma, Giovanni Pezzulo, and Alberto Corigliano. At its core lies active inference – a neuroscience-inspired Bayesian framework that integrates probabilistic reasoning and predictive modeling with the overarching goal of minimizing free energy. By conceptualizing the evolution of the active digital twin as a partially observable Markov decision process, the active inference agent continuously refines its generative model through Bayesian updates to forecast future states and observations. This proactive approach to decision-making strikes a delicate balance between pragmatic exploitation (maximizing goal-directed utility) and epistemic exploration (actively resolving uncertainty). The key innovation of the active digital twin paradigm is its ability to dynamically plan actions that minimize expected free energy. This metric encapsulates both the discrepancy between predicted and preferred future observations as well as the informational value gained from uncovering hidden states. As a result, active digital twins exhibit heightened levels of autonomy and resilience compared to their passive counterparts, offering superior capabilities for spontaneous exploration. To demonstrate the efficacy of this framework in practice,and of a railway bridge were assessed by the authors. Through this case study,showcase how they can revolutionize asset management by proactively identifying potential issues before they escalate into costly problems. In summary, the introduction of via represents a significant advancement in the field of engineering and applied sciences. By harnessing cutting-edge techniques inspired by neuroscience and Bayesian reasoning, this paradigm promises to usher in a new era of adaptive and intelligent systems capable of navigating complex and ever-changing environments with unparalleled efficiency.