Neuro-evolution in Neuroscience & Robotics: Insights from GECCO 2021 🧠🤖

Neuro-evolution at Work in Neuroscience and Robotics (wkspk110, WS - NEvo@Work) Eric O. Scott, Kenneth A. De Jong This presentation gives an overview of a unique multi-year research project in which neuro- evolution is being used to simultaneously develop brain models capable of replicating the neural activity of rats exploring mazes for rewards and simultaneously using those results to embed and evolve neuro-controllers in robots solving similar maze problems. On the neuroscience side, evolved brain-based recurrent spiking neural networks have been shown to be capable of modeling the spatial and working memory elements necessary for rats solving maze problems. On the robotics side, evolved neuro-controllers have been shown to be capable of generating comparable maze-solving behavior in robots. We discuss how these results fit into the field's broader ambition of using neuroscience to inform and enable the design of low-cost, low-power embedded systems for robotics and AI. Finally, there will be a short discussion of work in progress including the introduction of some co-evolutionary ideas. GECCO 2021 The Genetic and Evolutionary Computation Conference July 10-14, 2021 — Lille, France (online) https://gecco-2021.sigevo.org