Hybrid Incremental Neuroevolution of Planning and Locomotion (2022)
An update on the ECAL 2015 work described below has a page of its own.
Reactive and Deliberative 3D Virtual Creatures (ECAL 2015)
In this work we investigated how to go about building a system that generates complex adaptive behaviour in 3D virtual creatures.
Complexification Strategies for Incremental Evolution (ECAL 2013)
This work looked at strategies for presenting sub-components of a complex task to an evolving population such that the population could learn to generalise from the subtasks to the general solution. It turns out that the naive strategies of incrementally making the subtasks harder, or randomly presenting the simpler components, is not optimum in some situations. A more considered approach was shown to markedly improve the performance of the system.
This brief experiment played with the idea of using balance as a proxy for energy minimisation in walking agents.