The Free Energy Principle (FEP) and Active Inference represent a promising framework to model motor control. The FEP stems from Helmholtz's proposal that the brain's cognitive processes are not the result of direct elaboration of sensory data; rather, they are the outcome of probabilistic inference about the hidden causes causing the perceived sensations. Starting from classical Bayesian inference, Variational Free Energy (VFE) can be derived as a tractable quantity to be minimized in order to maintain homeostasis. The Active Inference framework suggests that such minimization can occur both through perceptual adaptation and action. Such framework is especially promising since it is able to model specific behaviors that remain widely unaddressed by other motor control theories; specifically, subtle movements are observed in the presence of multisensory conflict, a condition in which different sensory channels provide conflicting information about the environment. An example is represented by the Rubber Hand Illusion, where subjects experience a sense of body ownership over a rubber arm that is placed in a plausible configuration near their real arm (which is concealed from view). During such experiment, subjects perform subtle movements of the real arm towards the fake limb; these movements have the underlying goal of resolving the multisensory conflict arising from their proprioceptive and visual channels, which provide contrasting sensorial information about the arm configuration. The aims of this work are to provide a clear theoretical and analytical description of the Free Energy Principle; moreover, a simplified simulation environment is executed to demonstrate the feasibility of the active inference paradigm both during standard goal-oriented task and while experiencing multisensory conflicts.
Il Principio di Energia Libera (FEP) e l'Active Inference costituiscono un framework promettente per la modellazione del controllo motorio. Basandosi sulla prospettiva di Helmholtz, il FEP suggerisce che i processi cognitivi non derivino da una semplice elaborazione di dati sensoriali, ma siano il risultato di un'inferenza probabilistica sulle cause nascoste che generano le percezioni. Partendo dall'inferenza bayesiana classica, l'Energia Libera Variazionale (VFE) viene definita come una quantità trattabile che l'organismo minimizza per mantenere l'omeostasi. L'Active Inference propone che tale minimizzazione avvenga sia attraverso l'adattamento percettivo che tramite l'azione. Questo approccio è particolarmente rilevante poiché spiega comportamenti spesso ignorati da altre teorie del controllo motorio, come i micro-movimenti che emergono in presenza di conflitti multisensoriali. Un esempio emblematico è la Rubber Hand Illusion (RHI), in cui i soggetti percepiscono come propria una mano artificiale posta vicino al braccio reale (nascosto alla vista). In questo scenario, si osservano piccoli spostamenti del braccio reale verso quello finto: movimenti che mirano intrinsecamente a risolvere il conflitto tra il canale propriocettivo e quello visivo. L'obiettivo di questa tesi è fornire una rigorosa descrizione teorica e analitica del FEP. Inoltre, viene presentato e implementato un modello di simulazione semplificato per dimostrare l'efficacia del paradigma dell'Active Inference, sia nell'esecuzione di compiti orientati a un obiettivo (goal-oriented), sia durante l'esperienza di conflitti multisensoriali. Parole chiave: Principio di Energia Libera, Active Inference, conflitto multisensoriale, Rubber Hand Illusion.
Active inference: modelling motor control under multisensory conflict
ARRIGONI, ALESSANDRO
2024/2025
Abstract
The Free Energy Principle (FEP) and Active Inference represent a promising framework to model motor control. The FEP stems from Helmholtz's proposal that the brain's cognitive processes are not the result of direct elaboration of sensory data; rather, they are the outcome of probabilistic inference about the hidden causes causing the perceived sensations. Starting from classical Bayesian inference, Variational Free Energy (VFE) can be derived as a tractable quantity to be minimized in order to maintain homeostasis. The Active Inference framework suggests that such minimization can occur both through perceptual adaptation and action. Such framework is especially promising since it is able to model specific behaviors that remain widely unaddressed by other motor control theories; specifically, subtle movements are observed in the presence of multisensory conflict, a condition in which different sensory channels provide conflicting information about the environment. An example is represented by the Rubber Hand Illusion, where subjects experience a sense of body ownership over a rubber arm that is placed in a plausible configuration near their real arm (which is concealed from view). During such experiment, subjects perform subtle movements of the real arm towards the fake limb; these movements have the underlying goal of resolving the multisensory conflict arising from their proprioceptive and visual channels, which provide contrasting sensorial information about the arm configuration. The aims of this work are to provide a clear theoretical and analytical description of the Free Energy Principle; moreover, a simplified simulation environment is executed to demonstrate the feasibility of the active inference paradigm both during standard goal-oriented task and while experiencing multisensory conflicts.| File | Dimensione | Formato | |
|---|---|---|---|
|
Master_thesis.pdf
accessibile in internet per tutti
Dimensione
2.39 MB
Formato
Adobe PDF
|
2.39 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/253570