Abstract-This paper presents the framework, rules, games, controllers and results of the first General Video Game Playing Competition, held at the IEEE Conference on Computational Intelligence and Games in 2014. The competition proposes the challenge of creating controllers for general video game play, where a single agent must be able to play many different games, some of them unknown to the participants at the time of submitting their entries. This test can be seen as an approximation of General Artificial Intelligence, as the amount of game-dependent heuristics needs to be severely limited.The games employed are stochastic real-time scenarios (where the time budget to provide the next action is measured in milliseconds) with different winning conditions, scoring mechanisms, sprite types and available actions for the player. It is a responsibility of the agents to discover the mechanics of each game, the requirements to obtain a high score and the requisites to finally achieve victory. This paper describes all controllers submitted to the competition, with an in-depth description of four of them by their authors, including the winner and the runner-up entries of the contest. The paper also analyzes the performance of the different approaches submitted, and finally proposes future tracks for the competition.

Abstract. Upper Confidence Trees are a very efficient tool for solving Markov Decision Processes; originating in difficult games like the game of Go, it is in particular surprisingly efficient in high dimensional problems. It is known that it can be adapted to continuous domains in some cases (in particular continuous action spaces). We here present an extension of Upper Confidence Trees to continuous stochastic problems. We (i) show a deceptive problem on which the classical Upper Confidence Tree approach does not work, even with arbitrarily large computational power and with progressive widening (ii) propose an improvement, termed double-progressive widening, which takes care of the compromise between variance (we want infinitely many simulations for each action/state) and bias (we want sufficiently many nodes to avoid a bias by the first nodes) and which extends the classical progressive widening (iii) discuss its consistency and show experimentally that it performs well on the deceptive problem and on experimental benchmarks. We guess that the double-progressive widening trick can be used for other algorithms as well, as a general tool for ensuring a good bias/variance compromise in search algorithms.

Abstract-This paper showcases the setting and results of the first Two-Player General Video Game AI competition, which ran in 2016 at the IEEE World Congress on Computational Intelligence and the IEEE Conference on Computational Intelligence and Games. The challenges for the general game AI agents are expanded in this track from the single-player version, looking at direct player interaction in both competitive and cooperative environments of various types and degrees of difficulty. The focus is on the agents not only handling multiple problems, but also having to account for another intelligent entity in the game, who is expected to work towards their own goals (winning the game). This other player will possibly interact with first agent in a more engaging way than the environment or any non-playing character may do. The top competition entries are analyzed in detail and the performance of all agents is compared across the four sets of games. The results validate the competition system in assessing generality, as well as showing Monte Carlo Tree Search continuing to dominate by winning the overall Championship. However, this approach is closely followed by Rolling Horizon Evolutionary Algorithms, employed by the winner of the second leg of the contest.

In the Bayesian Reinforcement Learning (BRL) setting, agents try to maximise the collected rewards while interacting with their environment while using some prior knowledge that is accessed beforehand. Many BRL algorithms have already been proposed, but the benchmarks used to compare them are only relevant for specific cases. The paper addresses this problem, and provides a new BRL comparison methodology along with the corresponding open source library. In this methodology, a comparison criterion that measures the performance of algorithms on large sets of Markov Decision Processes (MDPs) drawn from some probability distributions is defined. In order to enable the comparison of non-anytime algorithms, our methodology also includes a detailed analysis of the computation time requirement of each algorithm. Our library is released with all source code and documentation: it includes three test problems, each of which has two different prior distributions, and seven state-of-the-art RL algorithms. Finally, our library is illustrated by comparing all the available algorithms and the results are discussed.

Abstract-Estimating the belief state is the main issue in games with Partial Observation. It is commonly done by heuristic methods, with no mathematical guarantee. We here focus on mathematically consistent belief state estimation methods, in the case of one-player games. We clearly separate the search algorithm (which might be e.g. alpha-beta or Monte-Carlo Tree Search) and the belief state estimation. We basically propose rejection methods and simple Monte-Carlo Markov Chain methods, with a time budget proportional to the time spent by the search algorithm on the situation at which the belief state is to be estimated; this is conveniently approximated by the number of simulations in the current node. While the approach is intended to be generic, we perform experiments on the wellknown Mines game, available on most Windows and Linux distributions. Interestingly, it detects non-trivial facts, e.g. the fact that the probability of winning the game is not the same for different moves, even those with the same probability of immediate death. The rejection method, which is slow but has no parameter and which is consistent in a non-asymptotic setting, performed better than the MCMC method in spite of tuning efforts.

Abstract. In the standard version of the UCT algorithm, in the case of a continuous set of decisions, the exploration of new decisions is done through blind search. This can lead to very inefficient exploration, particularly in the case of large dimension problems, which often happens in energy management problems, for instance. In an attempt to use the information gathered through past simulations to better explore new decisions, we propose a method named Blind Value (BV). It only requires the access to a function that randomly draws feasible decisions. We also implement it and compare it to the original version of continuous UCT. Our results show that it gives a significant increase in convergence speed, in dimensions 12 and 80.

Abstract:Bayesian Reinforcement Learning (BRL) agents aim to maximise the expected collected rewards obtained when interacting with an unknown Markov Decision Process (MDP) while using some prior knowledge. State-of-the-art BRL agents rely on frequent updates of the belief on the MDP, as new observations of the environment are made. This offers theoretical guarantees to converge to an optimum, but is computationally intractable, even on small-scale problems. In this paper, we present a method that circumvents this issue by training a parametric policy able to recommend an action directly from raw observations. Artificial Neural Networks (ANNs) are used to represent this policy, and are trained on the trajectories sampled from the prior. The trained model is then used online, and is able to act on the real MDP at a very low computational cost.Our new algorithm shows strong empirical performance, on a wide range of test problems, and is robust to inaccuracies of the prior distribution.

Abstract. We experiment the introduction of machine learning tools to improve Monte-Carlo Tree Search. More precisely, we propose the use of Direct Policy Search, a classical reinforcement learning paradigm, to learn the Monte-Carlo Move Generator. We experiment our algorithm on different forms of unit commitment problems, including experiments on a problem with both macrolevel and microlevel decisions.