Robotic AI learns to be spontaneous

Researchers use dynamic systems and machine learning to add spontaneity to artificial intelligence.

Autonomous functions for robots, such as spontaneity, are highly sought after. Many control mechanisms for autonomous robots are inspired by the functions of animals, including humans. Robotists often design robot behaviors using predefined control modules and methodologies, which makes them specific to the task, limiting their flexibility. Researchers offer an alternative method based on machine learning to design spontaneous behaviors by capitalizing on complex temporal patterns, such as the neural activities of the animal brain. They hope to see their design implemented on robotic platforms to enhance their autonomous capabilities.

Robots and their control software can be classified as a dynamic system, a mathematical model that describes the ever-changing internal states of something. There is a class of dynamic system called high-dimensional chaos, which has attracted many researchers as a powerful way to model the brains of animals. However, it is generally difficult to gain control over high-dimensional chaos due to the complexity of the system parameters and its sensitivity to changing initial conditions, a phenomenon popularized by the term “butterfly effect”. Researchers from the Intelligent Systems and Informatics Laboratory and the Next Generation Artificial Intelligence Research Center at the University of Tokyo are exploring new ways to harness the dynamics of high-dimensional chaos to implement human-like cognitive functions.

“There is an aspect of high-dimensional chaos called chaotic itinerancy (CI) that can explain brain activity during memory recall and association,” said PhD student Katsuma Inoue. “In robotics, CI has been a key tool for implementing spontaneous behavioral models. In this study, we propose a recipe for implementing CI in a simple and systematic way using only complicated time series models generated by high-dimensional chaos. We felt our approach has the potential for more robust and versatile applications when it comes to designing cognitive architectures. It allows us to design spontaneous behaviors without predefined explicit structures in the controller, which would otherwise serve as an obstacle. “

Reservoir Computing (RC) is a machine learning technique that is based on dynamic systems theory and provides the foundation for the team’s approach. RC is used to control a type of neural network called a recurrent neural network (RNN). Unlike other machine learning approaches that optimize all neural connections within a neural network, RC modifies only a few parameters while keeping all other connections of an RNN fixed, which makes it possible to train the system faster. When the researchers applied the principles of RC to a chaotic RNN, it showed the kind of spontaneous behavioral patterns they were hoping for. This has long proved to be a challenging task in the field of robotics and artificial intelligence. In addition, training for the network takes place before execution and in a short amount of time.

“Animal brains produce high-dimensional chaos in their activities, but how and why they use chaos remains unexplained. Our proposed model could offer insight into how chaos contributes to information processing in our brains, “said Professor Kohei Nakajima.” Furthermore, our recipe would have a wider impact outside the neuroscience field as it can potentially be applied to other chaotic systems as well. For example, next-generation neuromorphic devices inspired by biological neurons potentially show high-dimensional chaos and would be excellent candidates for implementing our recipe. I hope we will soon see artificial implementations of brain functions. “

Reference: “Designing spontaneous behavioral switching via chaotic itinerancy” by Katsuma Inoue, Kohei Nakajima and Yasuo Kuniyoshi, 11 November 2020, Advances in science.
DOI: 10.1126 / sciadv.abb3989

Financing:

This work was based on the results obtained from a project commissioned by the New Energy and Industrial Technology Development Organization (NEDO). KI was supported by JSPS KAKENHI (license number JP20J12815). KN was supported by JSPS KAKENHI (license number JP18H05472) and by MEXT Quantum Leap Flagship Program (MEXT Q-LEAP) (license number JPMXS0120319794). This work was supported by NEDO [serial numbers 15101156-0 (dated 24 June 2016) and 18101806-0 (dated 5 September 2018)] and president for Frontier AI Education, School of Information Science and Technology, and Next Generation AI Research Center [serial number not applicable (dated 1 June 2016)].