Towards open and expandable cognitive AI architectures for large-scale multi-agent human-robot collaborative learning

Learning from Demonstration (LfD) constitutes one of the most robust\nmethodologies for constructing efficient cognitive robotic systems. Despite the\nlarge body of research works already reported, current key technological\nchallenges include those of multi-agent learning and long-term autonomy.\nTowards this direction, a novel cognitive architecture for multi-agent LfD\nrobotic learning is introduced, targeting to enable the reliable deployment of\nopen, scalable and expandable robotic systems in large-scale and complex\nenvironments. In particular, the designed architecture capitalizes on the\nrecent advances in the Artificial Intelligence (AI) field, by establishing a\nFederated Learning (FL)-based framework for incarnating a multi-human\nmulti-robot collaborative learning environment. The fundamental\nconceptualization relies on employing multiple AI-empowered cognitive processes\n(implementing various robotic tasks) that operate at the edge nodes of a\nnetwork of robotic platforms, while global AI models (underpinning the\naforementioned robotic tasks) are collectively created and shared among the\nnetwork, by elegantly combining information from a large number of human-robot\ninteraction instances. Regarding pivotal novelties, the designed cognitive\narchitecture a) introduces a new FL-based formalism that extends the\nconventional LfD learning paradigm to support large-scale multi-agent\noperational settings, b) elaborates previous FL-based self-learning robotic\nschemes so as to incorporate the human in the learning loop and c) consolidates\nthe fundamental principles of FL with additional sophisticated AI-enabled\nlearning methodologies for modelling the multi-level inter-dependencies among\nthe robotic tasks. The applicability of the proposed framework is explained\nusing an example of a real-world industrial case study for agile\nproduction-based Critical Raw Materials (CRM) recovery.\n