Researchers and roboticists are continually trying to achieve autonomous functions in robots, and they often look toward the animal brain as a point of inspiration for control mechanisms. Because of the task-specific nature of robotic behavior, due to the reliance on predefined modules and control methodologies, they are often limited in flexibility.
The newest development in this area is coming out of the University of Tokyo, where researchers have created an alternative machine learning-based method to give robotic AI spontaneous behaviors. The team did this by relying on intricate temporal patterns, such as an animal brain's neural activities.
The research was published in Science Advances, titled “Designing spontaneous behavioral switching via chaotic itinerancy.”
A dynamical system is a mathematical model of the ever-changing internal states of something, which describes robots and their control software. Researchers are especially focused on high-dimensional chaos, a class of dynamical systems, due to its impressive ability to model animal brains.
Because of the complexity and sensitivity to varying initial conditions, high-dimensional chaos is especially challenging to control. To advance the field and clear this hurdle, researchers from the Intelligent Systems and Informatics Laboratory, and the Next Generation Artificial Intelligence Research Center at the University of Tokyo, have developed new ways to use high-dimensional chaos to provide robots with cognitive functions similar to humans.
Katsuma Inoue is a doctoral student working on the research.
“There is an aspect of high-dimensional chaos called chaotic itinerancy (CI) which can explain brain activity during memory recall and association,” Inoue said. “In robotics, CI has been a key tool for implementing spontaneous behavioral patterns. In this study, we propose a recipe for implementing CI in a simple and systematic fashion only using complicated time-series patterns generated by high-dimensional chaos. We felt our approach holds potential for more robust and versatile applications when it comes to designing cognitive architectures. It allows us to design spontaneous behaviors without any predefined explicit structures in the controller, which would otherwise serve as a hindrance.”
What is Reservoir Computing (RC)
The team relied heavily on reservoir computing (RC), a machine learning technique involving dynamical systems theory. RC is used to control recurrent neural networks (RNNs), and it keeps most connections of an RNN fixed while altering just a few parameters. This is different from other machine learning approaches, which often slightly alter all neural connections in a neural network, and it results in the system being able to be trained faster.
The researchers achieved the desired result when applying RC principles to a chaotic RNN, and it ended up demonstrating spontaneous behavioral patterns. The training for the network happens quickly and before execution.
“Animal brains yield high-dimensional chaos in their activities, but how and why they utilize chaos remains unexplained. Our proposed model could offer insight into how chaos contributes to information processing in our brains,” said Kohei Nakajima, an associate professor at the university. “Also, our recipe would have a broader impact outside the field of neuroscience since it can potentially be applied to other chaotic systems too. For example, next-generation neuromorphic devices inspired by biological neurons potentially exhibit high-dimensional chaos and would be excellent candidates for implementing our recipe. I hope we will see artificial implementations of brain functions before too long.”
The development is significant for the fields of robotics and artificial intelligence (AI), as researchers have been tackling this challenge for a while. It is the most recent example of how the fields are advancing at a fast pace.