For the 5th InternationalWorkshop on Active Inference (IWAI), we gathered in Oxford, UK on September 9–11, 2024. Around 130 active inference researchers from academia and industry attended the workshop in the beautiful Corpus Christi College, with a threeday program packed with two tutorials, six keynotes, a poster session, and seven sessions of paper presentations.

This volume presents the 17 full papers that were accepted and presented at the workshop. Out of 54 submissions, 17 full papers were selected through a double-blind review process. These papers cover a wide range of domains that find applications of active inference, ranging from robotics, decision-making and control, psychology, and representation learning, to theoretical advancements of learning and inference as well as active inference implementations. We also selected 33 contributions to be presented as posters.

The IWAI 2024 organizers would like to thank the Program Committee for their valuable review work, all authors for their contributions, Charel van Hoof and Sanjeev Namjoshi for their excellent tutorials, Emma Holmes, Chris Buckley, Rafal Bogacz, Pablo Lanillos, Ingmar Posner and Karl Friston for their inspiring keynotes, and of course all the attendees. We would also like to thank our sponsor VERSES AI, which made this event possible. [...]

Adaptation to external and internal changes is of major importance for robotic systems in uncertain environments. Here, we present a novel multisensory active inference (AIF) torque controller for industrial arms that shows how prediction can be used to resolve adaptation. Our controller, inspired by the predictive brain hypothesis, improves the capabilities of current AIF approaches by incorporating learning and multimodal integration of low- and high-dimensional sensor inputs (e.g., raw images) while simplifying the architecture. We performed a systematic evaluation of our model on a 7DoF Franka Emika Panda robot arm by comparing its behavior with previous AIF baselines and classic controllers, analyzing both qualitatively and quantitatively adaptation capabilities and control accuracy. The results showed improved control accuracy in goal-directed reaching with high noise rejection due to multimodal filtering, and adaptability to dynamical inertial changes, elasticity constraints, and human disturbances without the need to relearn the model or parameter retuning.

Computational models of visual attention in artificial intelligence and robotics have been inspired by the concept of a saliency map. These models account for the mutual information between the (current) visual information and its estimated causes. However, they fail to consider the circular causality between perception and action. In other words, they do not consider where to sample next, given current beliefs. Here, we reclaim salience as an active inference process that relies on two basic principles: uncertainty minimization and rhythmic scheduling. For this, we make a distinction between attention and salience. Briefly, we associate attention with precision control, i.e., the confidence with which beliefs can be updated given sampled sensory data, and salience with uncertainty minimization that underwrites the selection of future sensory data. Using this, we propose a new account of attention based on rhythmic precision-modulation and discuss its potential in robotics, providing numerical experiments that showcase its advantages for state and noise estimation, system identification and action selection for informative path planning