The problem of resource allocation of nonlinear networked control systems is investigated, where, unlike the well discussed case of triggering for stability, the objective is optimal triggering. An approximate dynamic programming approach is developed for solving problems with fixed final times initially and then it is extended to infinite horizon problems. Different cases including Zero-Order-Hold, Generalized Zero-Order-Hold, and stochastic networks are investigated. Afterwards, the developments are extended to the case of problems with unknown dynamics and a model-free scheme is presented for learning the (approximate) optimal solution. After detailed analyses of convergence, optimality, and stability of the results, the performance of the method is demonstrated through different numerical examples.

This paper studies two important signal processing aspects of equilibrium behavior in non-cooperative games arising in social networks, namely, reinforcement learning and detection of equilibrium play. The first part of the paper presents a reinforcement learning (adaptive filtering) algorithm that facilitates learning an equilibrium by resorting to diffusion cooperation strategies in a social network. Agents form homophilic social groups, within which they exchange past experiences over an undirected graph. It is shown that, if all agents follow the proposed algorithm, their global behavior is attracted to the correlated equilibria set of the game. The second part of the paper provides a test to detect if the actions of agents are consistent with play from the equilibrium of a concave potential game. The theory of revealed preference from microeconomics is used to construct a non-parametric decision test and statistical test which only require the probe and associated actions of agents. A stochastic gradient algorithm is given to optimize the probe in real time to minimize the Type-II error probabilities of the detection test subject to specified Type-I error probability. We provide a real-world example using the energy market, and a numerical example to detect malicious agents in an online social network.

Motivated by the recent applications of game-theoretical learning techniques to the design of distributed control systems, we study a class of control problems that can be formulated as potential games with continuous action sets, and we propose an actor-critic reinforcement learning algorithm that provably converges to equilibrium in this class of problems. The method employed is to analyse the learning process under study through a mean-field dynamical system that evolves in an infinite-dimensional function space (the space of probability distributions over the players' continuous controls). To do so, we extend the theory of finite-dimensional two-timescale stochastic approximation to an infinite-dimensional, Banach space setting, and we prove that the continuous dynamics of the process converge to equilibrium in the case of potential games. These results combine to give a provably-convergent learning algorithm in which players do not need to keep track of the controls selected by the other agents.

We study the model of projective simulation (PS), a novel approach to artificial intelligence based on stochastic processing of episodic memory which was recently introduced [H.J. Briegel and G. De las Cuevas. Sci. Rep. 2, 400, (2012)]. Here we provide a detailed analysis of the model and examine its performance, including its achievable efficiency, its learning times and the way both properties scale with the problems' dimension. In addition, we situate the PS agent in different learning scenarios, and study its learning abilities. A variety of new scenarios are being considered, thereby demonstrating the model's flexibility. Furthermore, to put the PS scheme in context, we compare its performance with those of Q-learning and learning classifier systems, two popular models in the field of reinforcement learning. It is shown that PS is a competitive artificial intelligence model of unique properties and strengths.

Conditional Value at Risk (CVaR) is a prominent risk measure that is being used extensively in various domains. We develop a new formula for the gradient of the CVaR in the form of a conditional expectation. Based on this formula, we propose a novel sampling-based estimator for the CVaR gradient, in the spirit of the likelihood-ratio method. We analyze the bias of the estimator, and prove the convergence of a corresponding stochastic gradient descent algorithm to a local CVaR optimum. Our method allows to consider CVaR optimization in new domains. As an example, we consider a reinforcement learning application, and learn a risk-sensitive controller for the game of Tetris.

We apply diffusion strategies to develop a fully-distributed cooperative reinforcement learning algorithm in which agents in a network communicate only with their immediate neighbors to improve predictions about their environment. The algorithm can also be applied to off-policy learning, meaning that the agents can predict the response to a behavior different from the actual policies they are following. The proposed distributed strategy is efficient, with linear complexity in both computation time and memory footprint. We provide a mean-square-error performance analysis and establish convergence under constant step-size updates, which endow the network with continuous learning capabilities. The results show a clear gain from cooperation: when the individual agents can estimate the solution, cooperation increases stability and reduces bias and variance of the prediction error; but, more importantly, the network is able to approach the optimal solution even when none of the individual agents can (e.g., when the individual behavior policies restrict each agent to sample a small portion of the state space).

Research on POMDP formulations for collaborative tasks in game AI applications (Nguyen et al. 2011; Macindoe, The development of new industrial robotic systems that operate Kaelbling, and Lozano-Pérez 2012; Silver and Veness in the same physical space as people highlights the 2010) also assumed a known human model. Additionally, emerging need for robots that can integrate seamlessly into previous partially observable formalisms (Ong et al. 2010; human group dynamics by adapting to the personalized style Bandyopadhyay et al. 2013; Broz, Nourbakhsh, and Simmons of human teammates. This adaptation requires learning a statistical 2011; Fern and Tadepalli 2010; Nguyen et al. 2011; model of human behavior and integrating this model Macindoe, Kaelbling, and Lozano-Pérez 2012) in assistive into the decision-making algorithm of the robot in a principled or collaborative tasks represented the preference or intention way. We present a framework for automatically learning of the human for their own actions, rather than those of human user models from joint-action demonstrations the robot, as the partially observable variable.

Engagement is a key factor in every social interaction, be it between humans or humans and robots. Many studies were aimed at designing robot behavior in order to sustain human engagement. Infants and children, however, learn how to engage their caregivers to receive more attention.We used a social robot platform, DragonBot, that learned which of its social behaviors retained human engagement. This was achieved by implementing a reinforcement learning algorithm, wherein the reward is the proximity and number of people near the robot. The experiment was run in the World Science Festival in New York, where hundreds of people interacted with the robot. After more than two continuous hours of interaction, the robot learned by itself that making a sad face was the most rewarding expression. Further analysis showed that after a sad face, people's engagement rose for thirty seconds. In other words, the robot learned by itself in two hours that almost no-one leaves a sad DragonBot.

In this work we present a method that allows to learn a cost function for motion planning of human-robot collaborative manipulation tasks where the human and the robot manipulate objects simultaneously in close proximity. Our approach is based on inverse optimal control which enables, considering a set of demonstrations, to find a cost function balancing different features. The cost function that is recovered from the human demonstrations is composed of elementary features, which are designed to encode notions such as safely, legibility and efficiency of the manipulation motions. We demonstrate the approach on data gathered from motion capture of human-human manipulation in close proximity of blocks on a table. To demonstrate the feasibility and efficacy of our approach we provide initial test results consisting of learning a cost function and then planning for the human kinematic model used in the learning phase.

An agent continuously performing different tasks in the same domain has the opportunity to learn, over the course of its operational lifetime, about the behavioural regularities afforded by the domain. This paper addresses the problem of learning a task independent behaviour model based on the underlying structure of a domain which is common across multiple tasks presented to an autonomous agent. Our approach involves learning action priors: a behavioural model which encodes a notion of local common sense behaviours in the domain, conditioned on either the state or observations of the agent. This knowledge is accumulated and transferred as an exploration behaviour whenever a new task is presented to the agent. The effect is that as the agent encounters more tasks, it is able to learn them faster and achieve greater overall performance. This approach is illustrated in experiments in a simulated extended navigation domain.