We propose a developmental process for high variety low volume production in the context of an intelligent multi-robot production system involving human agents in different roles. The goal is to provide a hierarchical framework, where observation and actions of multiple agents, i.e. robots and human agents (e.g. factory personnel and customers), are presented as collaborative plans with synchronized actions of the agents. We propose a learning framework based on hierarchical active inference, utilizing machine learning methods, like Markov models and deep neural networks, which enables persistent learning of completely new human-robot and robot-robot collaboration skills and improvement of existing human-robot and robot-robot collaboration skills through continuous monitoring and optimization of operations.

2023

2026