We have been supported by the Russian Science Foundation

Investigation of the nonlinear dynamics of neural networks for problems of cognitive navigation (project 15-12-10018: 2015-2017)

This project aims at the development of integrative systems and technologies to solve the problem of navigation in social environments. We study artificial neural networks serving as a prototype of a "cognitive core" capable of controlling the navigation of mobile robots.

Modern navigation strategies implicitly assume that the robots have to move among people without “disturbing” them. Nevertheless, a humanoid robot may induce people’s cooperation. Then we can formulate the problem of recursive cognition: Actions of a humanoid robot depend on the actions of pedestrians, which in turn depend on the robot’s actions and so on.

Recently, it has been shown that animals for orientation in space use cognitive maps. However, generalization of this concept to dynamic situations and especially to recursive cognition has significant difficulties. Earlier we suggested that the brain for internal representation of spatiotemporal situations does not use the time dimension explicitly. Instead, cognition emerges from the transformation of dynamic situations into static cognitive maps or, so-called, compact internal representations (CIR).

We propose, on the one hand, to investigate experimentally the neural structures involved in generation of CIRs and, on the other hand, to generalize the CIR theory and to study the processes of cognitive cooperation. The developed artificial neural network will recursively "encapsulate" the dynamics of other cognitive processes. To solve this problem the project combines a wide range of interdisciplinary methods, including conceptual modeling, recording of the neural activity in-vivo and in-vitro, computer simulations in-silico, and experimental verification using mobile robots.

Proposed interdisciplinary approach has no competitors, which supports our expectation of receiving pioneering results in the development of artificial intelligent systems. Ideally, neural architecture should emulate some of the computational capacities of the brain. In the future, the proposed cognitive core can be used to create commercial prototypes of intelligent robots and "smart" prostheses. Such prostheses will have cognitive abilities, which will allow reducing the user control over the prosthesis, an important condition for improving the life quality of a large group of people with disabilities.