Abstract: Recent advances in technology both in hardware and soft-ware have fuelled visions of swarms of robots being sent to remote or hazardous environments in which they will need to survive over long periods of time. Environments will be unknown and potentially dynamic, requiring autonomous adaptation by the swarm. Ensuring the integrity of a robot swarm in terms of maintaining a stable population of functioning robots over long periods of time in unknown environments is a mandatory prerequisite for building more complex systems that achieve user-defined tasks. In this talk, I will summarise a broad range of work that stems from using “Environment-driven Distributed Evolutionary Adaptation Algorithms” to evolve swarms in complex environments. In particular, we build on an algorithm called mEDEA, first introduced by Bredeche and Montanier in 2012, which has no central control, and relies on interactions between individual robots to spread and evolve controllers.
I will begin by discussing how a purely environment-driven approach to evolution can be augmented with locally-derived fitness information, which is used to influence evolution, and go on to consider how individual adaptation mechanisms can be mixed with evolution to enhance performance. Finally, I will describe recent work in which mEDEA is enhanced with a diversity maintenance mechanism based on an illumination algorithm (Map-Elites) to evolve a population of behaviourally diverse robots. As an aside, I will also present a method that uses surface-plots to explore the relationship between parameter setting and emergent behaviour and enables informed selection of parameters for experiments.
Biography: Prof. Emma Hart directs the Centre for Algorithms, Visualisation and Evolutionary Robotics at Edinburgh Napier University, and leads the newly established Evolutionary Swarm Robotics Laboratory. She is Editor-in-Chief of the journal Evolutionary Computation and play an active role in the Evolutionary Computation community: she was General Chair of PPSN in Edinburgh in 2016 and Track Chair @ GECCO for the past two years for the Complex and Adaptive Systems track. Her research spans a broad range of topics, from Evolutionary Robotics to Optimisation, with a particular focus on ensemble methods for optimisation and learning, and on lifelong learning – systems that continue to adapt and evolve over time as they gain experience. Her work has been funded by EPSRC and Leverhulme and includes a recently funded 4 year EPSRC project “Autonomous Robot Evolution: from Cradle to Grave” that will simultaneously evolve both morphology and controllers of robotics within a hybrid software-hardware architecture, exploiting the latest advances in 3D printing.