There are 6 repositories under active-inference topic.
Deep active inference agents using Monte-Carlo methods
PID-like control implemented as active inference with linear generative models
Implementation/simulation of active neural generative coding (ANGC) for training neurobiologically-plausible active inference agent models.
Active Inference & Category Theory
Homing Piegon is an inference framework implementing Variational Message Passing. It can be used to implement an Active Inference agent that performs planning using a Tree Search algorithm that can been seen as a form of Bayesian Model Expansion.
Active inference agent and corresponding environment in Unity used in the study "A deep active inference model of the rubber-hand illusion"
Implementation of the paper "A neural active inference model of perceptual-motor learning" published on Computational Neuroscience in 2023.
[NeurIPS 2021] Contrastive learning formulation of the active inference framework, for matching visual goal states.
Archive of active inference agents based on reactive message passing.
A Bayesian cruise controller. A minimal model of velocity regulation for a block on a frictionless surface.
Code, figures, animations for a NARX-EFE based agent.
Deep Active Inference of Mountain Car Problem
This repository focused on enhancing the Active Inference Controller (AIC) for robotic manipulation tasks on an Interbotix PincherX 150 5-DOF Robot. A modified control framework called the Reactive Active Inference Controller (ReAIC) outperfors the AIC in adaptation experiments.
Playground for active inference in Python
Book chapter - dev code
Application of active inference to reinforcement learning.
Design of an acrobot stabilization controller using active inference.
Design of an oscillator position controller using active inference.
Multi-Agent Robot Learning algorithm using Deep Active Inference (DAI) for road hazard anomaly detection and Soft Actor Critic decomposed for multi-agent settings (mSAC)
This repository contains all the code relevant to my PhD research (2019-2023) at University of Amsterdam, funded by a Research Talent Grant from Netherlands Organisation for Scientific Research (NWO). Pursuant to the Open Access requirements of the NWO, all code here is published on the basis of an MIT licence or equivalent.
Personal website of Leo D'Amato
Minimal model of tool discovery and tool innovation using active inference
Sophisticated Learning Implementations: MATLAB and Python code for the "Sophisticated Learning" algorithm from Hodson et al.'s research on model-based planning.