There are widely sought after autonomous roles for robots, such as spontaneity. The functions of animals including humans, motivate many control mechanisms for autonomous robots. Using predefined modules and control methodologies, roboticists frequently design robot behaviors, which makes them task-specific, restricting their versatility. Researchers propose an alternative approach based on machine learning for designing random behaviors by capitalizing on complex temporal patterns, such as animal brain neural activities. To strengthen their autonomous capabilities, they expect to see their concept implemented in robotic platforms.
It is possible to identify robots and their control software as a dynamical structure, a mathematical model that describes something’s ever-changing internal states. There is a class of high-dimensional chaos called a dynamic structure, which has attracted many researchers as it is a powerful way to model animal brains. Nevertheless, due to the ambiguity of the system parameters and its vulnerability to varying initial conditions, a phenomenon popularized by the term “butterfly effect,” it is generally difficult to gain control over high-dimensional chaos. Researchers from the Intelligent Systems and Informatics Laboratory and the University of Tokyo’s Next Generation Artificial Intelligence Research Center do not explore any.
“There is an element of high-dimensional chaos called chaotic itinerancy (CI) which during memory recall and association, can explain brain activity,” said doctoral student Katsuma Inoue. CI was a key tool for the introduction of random behavioral patterns in robotics. In this research, we propose a recipe for the easy and systematic implementation of CI only using complicated time-series patterns created by high-dimensional chaos. We thought that when it comes to designing cognitive architectures, our method has the potential for more stable and scalable applications.
Reservoir computing (RC) is a technique of machine learning that builds on the principle of dynamical systems and provides the basis for the methodology of the team. A type of neural network called a recurrent neural network (RNN) is managed using RC. RC just changes certain parameters while keeping all other RNN connections unchanged, unlike other machine learning methods that tune all neural connections within a neural network, which makes it easier to train the system faster. It demonstrated the kind of random behavioral patterns they were looking for when the researchers applied RC concepts to a chaotic RNN. This has proved to be a daunting challenge in the world of robotics and artificial intelligence for some time now. In addition, before execution and in a limited period of time, preparation for the network takes place.
“In their operations, animal brains create high-dimensional chaos, but it remains unclear how and why they use chaos. Our proposed model may give insight into how chaos leads to the processing of knowledge in our brains,” said Associate Professor Kohei Nakajima. Our recipe will also have a broader effect beyond the world of neuroscience as it may theoretically also be extended to other chaotic structures. For example, biological neuron-inspired neuromorphic devices of the next decade will demonstrate high-dimensional chaos which will be ideal candidates for our recipe to be introduced. I think we can see artificial brain function implementations.
Source of Story: Provision of supplies from the University of Tokyo. Note: For style and length, material can be edited.
No Comments
Leave a comment Cancel