Developing autonomous cognitive-developmental robots is one of the dreams of robotics. An autonomous cognitive system should be able to learn and adapt to its environment through interactions. Importantly, the experience is based on its sensorimotor systems. Creating cognitive dynamics that allow a robot to develop and learn based on the robot’s own action and perception cycles is a critical challenge in cognitive and developmental robotics. The autonomous learning process that occurs throughout development is also referred to as lifelong learning, and it is thought to be the foundation for the development of social capabilities necessary for adaptive collaborative robots.
Based on outstanding success in deep learning and probabilistic generative models in the 2010s, world models are attracting attention in artificial intelligence. A cognitive system (e.g., an agent) that learns a world model, with itself included, will be capable of predicting its future sensory observations and to optimize its controller (i.e., behavior) based on the prediction of the sensory consequences of its actions. The idea is closely related to predictive coding that has been studied in neurorobotics to develop neuro-dynamics realizing adaptive behaviors and social perception. Predictive coding and world models also share the same fundamental idea with the free-energy principle which is an influential theory in neuroscience nowadays.
The world model-based approach is promising. However, the many applications and studies of world models tend to be limited to simulation studies. The problems and challenges of developing autonomous cognitive-developmental robots based on world models, predictive coding, and the free-energy principle have not been fully explored on real and situated robots. These approaches are based on a generative view of cognition. In studies about cognitive development and symbol emergence in robotics, many computational cognitive models based on probabilistic generative models have been developed.
This study is focussing on theoretical and practical studies that enable us to create a real-world embodied cognitive systems based on world model-based approach.
Karl Friston, Rosalyn J. Moran, Yukie Nagai, Tadahiro Taniguchi, Hiroaki Gomi, Josh Tenenbaum, World model learning and inference, Neural Networks, 144(-), 573-590, 2021. https://doi.org/10.1016/j.neunet.2021.09.011
Masashi Okada, Tadahiro Taniguchi, Variational Inference MPC for Bayesian Model-based Reinforcement Learning, Conference on Robot Learning (CoRL) , 2019, paper
Masashi Okada, Norio Kosaka, Tadahiro Taniguchi, PlaNet of the Bayesians: Reconsidering and Improving Deep Planning Network by Incorporating Bayesian Inference, IEEE International Conference on Intelligent Robots and Systems (IROS), 2020, paper
Masashi Okada, Tadahiro Taniguchi, Dreaming: Model-based Reinforcement Learning by Latent Imagination without Reconstruction, IEEE International Conference on Robotics and Automation (ICRA) , 2021, paper
Building a human-like integrative artificial cognitive system, that is, an artificial general intelligence (AGI), is the holy grail of the artificial intelligence (AI) field. Furthermore, a computational model that enables an artificial system to achieve cognitive development will be an excellent reference for brain and cognitive science. This paper describes an approach to developing a cognitive architecture by integrating elemental cognitive modules to enable the training of the modules as a whole. This approach is based on two ideas: (1) brain-inspired AI, learning human brain architecture to build human-level intelligence, and (2) a probabilistic generative model (PGM)-based cognitive architecture to develop a cognitive system for developmental robots by integrating PGMs. The proposed development framework is called a whole-brain PGM (WB-PGM), which differs fundamentally from existing cognitive architectures. It can learn continuously through a system based on sensory-motor information.
As PGMs describe explicit informational relationships between variables, WB-PGM provides interpretable guidance from computational sciences to brain science. By providing such information, researchers in neuroscience can provide feedback to researchers in AI and robotics on what the current models lack with reference to the brain. Further, it can facilitate collaboration among researchers in neuro-cognitive sciences as well as AI and robotics.