Search

We present an architecture where a feedback controller derived on an approximate model of the environment assists the learning process to enhance its data efficiency. This architecture, which we term as Control-Tutored Q-learning (CTQL), is presented in two alternative flavours. The former is based on defining the reward function so that a Boolean condition can be used to determine when the control tutor policy is adopted, while the latter, termed as probabilistic CTQL (pCTQL), is instead based on executing calls to the tutor with a certain probability during learning. Both approaches are validated, and thoroughly benchmarked against Q-Learning, by considering the stabilization of an inverted pendulum as defined in OpenAI Gym as a representative problem.

Understanding the mechanisms underlying the emergence of leadership in multi-agent systems is still under investigation in many areas of research where group coordination is involved. While leadership has been mostly investigated in the case of animal groups, only a few works address the problem of leadership emergence in human ensembles, e.g. pedestrian walking, group dance. In this paper we study the emergence of leadership in the specific scenario of a small walking group. Our aim is to unveil the main mechanisms emerging in a human group when leader or follower roles are not designated a priori. Two groups of participants were asked to walk together and turn or change speed at self-selected times. Data were analysed using time-dependent cross correlation to infer leader-follower interactions between each pair of group members. The results indicate that leadership emergence is due both to contextual factors, such as an individual's position in the group, and to personal factors, such as an individual's characteristic locomotor behaviour. Our approach can easily be extended to larger groups and other scenarios such as team sports and emergency evacuations.

We study dynamical systems that switch between two different vector fields depending on a discrete variable and with a delay. When the delay reaches a problem-dependent critical value so-called event collisions occur. This paper classifies and analyzes event collisions, a special type of discontinuity induced bifurcations, for periodic orbits. Our focus is on event collisions of symmetric periodic orbits in systems with full reflection symmetry, a symmetry that is prevalent in applications. We derive an implicit expression for the Poincare map near the colliding periodic orbit. The Poincare map is piecewise smooth, finite-dimensional, and changes the dimension of its image at the collision. In the second part of the paper we apply this general result to the class of unstable linear single-degree-of-freedom oscillators where we detect and continue numerically collisions of invariant tori. Moreover, we observe that attracting closed invariant polygons emerge at the torus collision., Comment: 28 pages

We study synchronization phenomena in scale-free networks of asymetrically coupled dynamical systems featuring degree-degree correlation in the connection wiring. We show that when the interaction is dominant from the high-degree to the low-degree or from the low-degree to the high-degree nodes, synchronization is enhanced for assortative or disassortative degree-degree correlations, Comment: 11 pages, 4 figures

In this paper we introduce a new model of data packet transport, based on a stochastic approach with the aim of characterizing the load distribution on complex networks. Moreover we analyze the load standard deviation as an index of uniformity of the distribution of packets within the network, to characterize the effects of the network topology. We measure such index on the model proposed by Watts and Strogatz as the redirection probability is increased. We find that the uniformity of the load spread is maximized in the intermediate region, at which the small world effect is observed and both global and local efficiency are high. Moreover we analyze the relationship between load centrality and degree centrality as an approximate measure of the load at the edges. Analogous results are obtained for the load variance computed at the edges as well as at the vertices., Comment: 6 pages, 5 figures. Included in conference proceedings International Conference PhysCon 2005 August 24-26, 2005, Saint Petersburg, RUSSIA