Embodied evolutionary robotics (EER) designates a set of methods and algorithms to design adaptive collective robotic systems, where each robot can communicate only with close-by robots (Watson et al., 2002)(Bredeche et al., 2012). Lately, it has become apparent that the algorithms used in EER are related to classical models of population genetics. A very interesting aspect of evolutionary robotics is that it can be used to simulate interactions between individual during the evolutionary process. We wish to explore further the path of designing a new method to characterize the evolution of a trait in both EER and biological evolution.
During this internship, we will use different flavors of EER to generate phylogenetic data (trees of ancestors) with respect to a species evolving to adapt to an uncontrolled environment. Using fixation trajectories of the genetic traits, we will explore whether one can assess whether the fixation of a trait occurred by chance (i.e. genetic drift) or by means of selection. In particular, we will consider the case where the whole trajectory is unknown but several reference points are available. This is a very common situation with biological data, where very few points (sometimes even just two points) are used to fit a classic population genetics equation of invasion.
From the evolutionary biology viewpoint, the goal will be to provide both a yes/no answer on the existence of a selection pressure, as well as a confidence bound in the positive case. This work would have application in on-going works on real-time evolution of small organisms such as the Escherichia coli bacteria as well as in paleontological data where only few points are available.
This new method will be tested on two kind of data: (1) artificial data, generated using a robotic simulation and (2) real data, from a long-term evolution in vitro experiment. We will also explore how to reproduce real data in simulation, in order to fit a model so as to better understand evolutionary dynamics at work with the real system.
Agenda:
- implement variations of simple embodied evolutionary algorithms in simulations to generate data
- characterise the algorithms with respect to their ability to fixate particular genetic traits (selection pressure)
- extend the existing tool by providing a likelihood estimation of observing fixation
- test with artificially generated data and with real data from in vitro experiments
- learn an artificial model from the data
<strong>Administrative information</strong>
The internship will be done at ISIR/UPMC, starting Feb. 1st for 6 months (internship gratification is approx. 500 euros per month).
Co-advisors are: Nicolas Bredeche (ISIR, UPMC), Guillaume Achaz (ABI, Sorbonne Univ., UPMC)
<strong>Short bibliography</strong>
Embodied evolution: Distributing an evolutionary algorithm in a population of robots
RA Watson, SG Ficici, JB Pollack
Robotics and Autonomous Systems 39 (1), 1-18
Environment-driven Embodied Evolution in a Population of Autonomous Agents
Bredeche N., J.-M. Montanier
Proc. of the 11th International Conference on Parallel Problem Solving From Nature (PPSN 2010)