Abstract: In evolutionary robotics, controllers are often represented as
networks. Modularity is a desirable trait of such networks
because modular networks are resistant to catastrophic forgetting
and tend to have less connections than nonmodular
ones. However, these advantages can only be realized if the
control task is solvable by a modular network, and for any
given practical task the control task depends on the choice of
the robot’s morphology. Here we provide an example of a
task solvable by robots with two different morphologies. We
consider the most extreme kind of modularity – disconnectedness
– and show that with the first morphology the task
can be solved by a disconnected controller with few connections.
On the other hand, the second morphology makes the
task provably impossible for disconnected controllers and requires
about three times more connections. For this morphology,
most controllers that partially solve the task constitute
local optima, forming an extremely deceptive fitness landscape.
We show empirically that in this case a connection
cost-based evolutionary algorithm for evolving modular controllers
is greatly slowed down compared to the first morphology’s
case. Finally, this performance gap increases as the task
is scaled up. These results show that the morphology may be
a major factor determining the performance of controller optimization.
Although in our task the optimal morphology is
obvious to a human designer, we hypothesize that as evolutionary
robotics is scaled to more sophisticated tasks the optimization
of morphology alongside the control might become
a requirement for evolving modular controllers.
[edit database entry]
Bongard's work focuses on understanding the general nature of cognition, regardless of whether it is found in humans, animals or robots. This unique approach focuses on the role that morphology and evolution plays in cognition. Addressing these questions has taken him into the fields of biology, psychology, engineering and computer science.
Danforth is an applied mathematician interested in modeling a variety of physical, biological, and social phenomenon. He has applied principles of chaos theory to improve weather forecasts as a member of the Mathematics and Climate Research Network, and developed a real-time remote sensor of global happiness using messages from Twitter: the Hedonometer. Danforth co-runs the Computational Story Lab with Peter Dodds, and helps run UVM's reading group on complexity.
Laurent studies the interaction of structure and dynamics. His research involves network theory, statistical physics and nonlinear dynamics along with their applications in epidemiology, ecology, biology, and sociology. Recent projects include comparing complex networks of different nature, the coevolution of human behavior and infectious diseases, understanding the role of forest shape in determining stability of tropical forests, as well as the impact of echo chambers in political discussions.
Hines' work broadly focuses on finding ways to make electric energy more reliable, more affordable, with less environmental impact. Particular topics of interest include understanding the mechanisms by which small problems in the power grid become large blackouts, identifying and mitigating the stresses caused by large amounts of electric vehicle charging, and quantifying the impact of high penetrations of wind/solar on electricity systems.
Bagrow's interests include: Complex Networks (community detection, social modeling and human dynamics, statistical phenomena, graph similarity and isomorphism), Statistical Physics (non-equilibrium methods, phase transitions, percolation, interacting particle systems, spin glasses), and Optimization(glassy techniques such as simulated/quantum annealing, (non-gradient) minimization of noisy objective functions).