Modeling Neural Development with Braitenberg Vehicles

Evolving embodied neural networks in Braitenberg vehicles to study brain development and connectome motifs

completed 2019 –2019
Embodied IntelligenceArtificial IntelligenceBiologyComplexityRobotics

A Google Summer of Code 2019 project hosted by INCF and carried out within the Orthogonal Research Lab and Representational Brains & Phenotypes community. Contributor: Stefan Dvoretskii. Mentor: Bradly Alicea.

Abstract

This project is about building an embodied cognitive simulation, i.e. that in which robots we call vehicles have a body and a simple “mind”, represented by a neural activational network. The body has a defined shape, activator sensors that capture signals from the environment, and motors, that move the vehicle as a reaction to the signals. We then evolve neural networks inside vehicles using a Genetic Algorithm with an appropriate fitness function, and hope to observe some natural behavioural patterns as well as certain connectome motifs seen in nature. This would allow us to reproduce the very same behaviours’ simulation, as well as hypothesize on correspondence of connectome motifs to specific behaviours. This project’s results have possible applications in brain development studies, as well as transferring synthesized behaviour models to robots and enriching virtual embodied systems’ intelligence (e.g. game AI).

Outcomes

The primary software outcome of this project is BraGenBrain, a genetic-algorithm-based simulation suite implementing the framework described above.

Resources