Bayesian role discovery for multi-agent reinforcement learning

This article shows that seemingly diverse implementations of multi-agent reinforcement learning share the same basic building block in their learning dynamics: a mathematical term that is closely related to the gradient of the expected reward. Specifically, two independent branches of multi-agent learning research can be distinguished based on their respective assumptions and premises. The first branch assumes that the value function of the game is known to all players, which is then used to update the learning policy based on Gradient Ascent. Notable algorithms in this branch include Infinitesimal Gradient Ascent (IGA) [7], the variation Win or Learn Fast IGA (WoLF) [3] and the Weighted Policy Learner [1]. The second branch of multi-agent learning is concerned with learning in unknown environments, using interaction-based Reinforcement Learning, and contains those algorithms which have been shown to be formally connected to the replicator equations of Evolutionary Game Theory. In this case, the learning agent updates its policy based on a sequence of pairs that indicate the quality of the actions taken. Notable algorithms include Cross Learning (CL) [4], Regret Minimization (RM) [6], and Frequency Adjusted Q-learning (FAQ) [5]. This article demonstrates inherent similarities between these diverse families of algorithms by comparing their underlying learning dynamics, derived as the continuous time limit of their policy updates. These dynamics have already been investigated for algorithms from each family separately [1, 2, 3, 5, 6, 7], however, they have not yet been discussed in context of the relation to each other, and the origin of their similarity has not been discussed satisfactorily. In addition to the formal derivation, directional field plots of the learning dynamics in representative classes of two-player two-action games illustrate the similarities and strengthen the theoretical findings.

This paper investigates a relatively new direction in Multiagent Reinforcement Learning. Most multiagent learning techniques focus on Nash equilibria as elements of both the learning algorithm and its evaluation criteria. In contrast, we propose a multiagent learning algorithm that is optimal in the sense of finding a best-response policy, rather than in reaching an equilibrium. We present the first learning algorithm that is provably optimal against restricted classes of non-stationary opponents. The algorithm infers an accurate model of the opponentýs non-stationary strategy, and simultaneously creates a best-response policy against that strategy. Our learning algorithm works within the very general framework of n-player, general-sum stochastic games, and learns both the game structure and its associated optimal policy.

Coordinated multi-agent reinforcement learning (MARL) provides a promising approach to scaling learning in large cooperative multi-agent systems. Distributed constraint optimization (DCOP) techniques have been used to coordinate action selection among agents during both the learning phase and the policy execution phase (if learning is off-line) to ensure good overall system performance. However, running DCOP algorithms for each action selection through the whole system results in significant communication among agents, which is not practical for most applications with limited communication bandwidth. In this paper, we develop a learning approach that generalizes previous coordinated MARL approaches that use DCOP algorithms and enables MARL to be conducted over a spectrum from independent learning (without communication) to fully coordinated learning depending on agents' communication bandwidth. Our approach defines an interaction measure that allows agents to dynamically identify their beneficial coordination set (i.e., whom to coordinate with) in different situations and to trade off its performance and communication cost. By limiting their coordination set, agents dynamically decompose the coordination network in a distributed way, resulting in dramatically reduced communication for DCOP algorithms without significantly affecting overall learning performance. Essentially, our learning approach conducts co-adaptation of agents' policy learning and coordination set identification, which outperforms approaches that sequence them.

Solving complex but structured problems in a decentralized manner via multiagent collaboration has received much attention in recent years. This is natural, as on one hand, multiagent systems usually possess a structure that determines the allowable interactions among the agents; and on the other hand, the single most pressing need in a cooperative multiagent system is to coordinate the local policies of autonomous agents with restricted capabilities to serve a system-wide goal. The presence of uncertainty makes this even more challenging, as the agents face the additional need to learn the unknown environment parameters while forming (and following) local policies in an online fashion. In this paper, we provide the first Bayesian reinforcement learning (BRL) approach for distributed coordination and learning in a cooperative multiagent system by devising two solutions to this type of problem. More specifically, we show how the Value of Perfect Information (VPI) can be used to perform efficient decentralised exploration in both model-based and model-free BRL, and in the latter case, provide a closed form solution for VPI, correcting a decade old result by Dearden, Friedman and Russell. To evaluate these solutions, we present experimental results comparing their relative merits, and demonstrate empirically that both solutions outperform an existing multiagent learning method, representative of the state-of-the-art.

Model-free episodic reinforcement learning problems define the environment reward with functions that often provide only sparse information throughout the task. Consequently, agents are not given enough feedback about the fitness of their actions until the task ends with success or failure. Previous work addresses this problem with reward shaping. In this paper we introduce a novel approach to improve model-free reinforcement learning agents' performance with a three step approach. Specifically, we collect demonstration data, use the data to recover a linear function using inverse reinforcement learning and we use the recovered function for potential-based reward shaping. Our approach is model-free and scalable to high dimensional domains. To show the scalability of our approach we present two sets of experiments in a two dimensional Maze domain, and the 27 dimensional Mario AI domain. We compare the performance of our algorithm to previously introduced reinforcement learning from demonstration algorithms. Our experiments show that our approach outperforms the state-of-the-art in cumulative reward, learning rate and asymptotic performance.

Highly accurate sensor measurements are crucial in order for power plants to effectively operate, as well as to predict and subsequently prevent any potentially catastrophic failures. As the cost of sensors decreases while their power increases, distributed sensor networks become a more attractive option for implementation in power plants. In this work, we investi- gate the use of a distributed sensor network to achieve highly accurate measurements. We apply shaped rewards to local components and use a simple learning algorithm at each sen- sor in order to maximize those rewards. Our results show that the measurements from a sensor network trained us- ing shaped rewards are up to two orders of magnitude more accurate than a sensor network trained with a traditional global reward. Further, the algorithm proposed scales well to large networks, and is robust to measurement noise and sensor failures.

Reinforcement Learning (RL) Markov Decision Processes is studied with an emphasis on the well-studied exploration problem. We first formulate and discuss a definition of “efficient” algorithms that is termed Probably Approximately Correct (PAC) in RL. Next we provide general sufficient conditions for such an algorithm that applies to several different modeling assumptions. The conditions can be used to demonstrate that efficient learning is possible in finite MDPs, with either a model-based or model-free approach, in factored MDPs, and in continuous MDPs with linear dynamics. In the reinforcement-learning (RL) problem (Sutton & Barto 1998), an agent acts in an unknown or incompletely known environment with the goal of maximizing an external reward signal. In the most standard mathematical formulation of the problem, the environment is modeled as a Markov Decision Process (MDP) and the goal of the agent is to obtain near-optimal discounted return. Over the years, many algorithms have been proposed for this problem, but analyses of their performances have been relatively scarce. In fact, until recently, most theoretical guarantees have been that certain algorithms will discover an optimal policy in the limit, after an infinite amount of experience. In contrast, several attempts have been made to study “Probably Approximately Correct” or PAC-MDP algorithms, which exhibit near-optimal behavior in polynomial time and experience. This paper discusses several extensions of those results. We present a theorem that provides sufficient conditions for an algorithm to be PAC-MDP. We examine these conditions and show how they can be applied to prove that efficient learning is possible in three interesting scenerios: finite MDPs (i.e. the “Tabular case”),

In this work, we proposed a novel Hybrid Reinforcement Learning algorithm that incorporates the model-based reinforcement learning scheme with the model-free reinforcement learning framework, where the sample efficiency was improved while still retaining the desired control performance. The proposed learning framework was successfully applied to a complex biped robot model, where the robot was able to maintain balance on a rotating platform. Experiment results also showed that the proposed algorithm was able to generate stable walking gaits for the robot to walk on both static and dynamic platforms without falling down.

In this paper, we investigate multi-agent learning (MAL) in a multi-agent resource selection problem (MARS) in which a large group of agents are competing for common resources. Since agents in such a setting are self-interested, MAL in MARS domains typically focuses on the convergence to a set of non-cooperative equilibria. As seen in the example of prisoner's dilemma, however, selfish equilibria are not necessarily optimal with respect to the natural objective function of a target problem, e.g., resource utilization in the case of MARS. Conversely, a centrally administered optimization of physically distributed agents is infeasible in many real-life applications such as transportation traffic problems. In order to explore the possibility for a middle ground solution, we analyze two types of costs for evaluating MAL algorithms in this context. The quality loss of a selfish algorithm can be quantitatively measured by the price of anarchy, i.e., the ratio of the objective function value of a selfish solution to that of an optimal solution. Analogously, we introduce the price of monarchy of a learning algorithm to quantify the practical cost of coordination in terms of communication cost. We then introduce a multi-agent social learning approach named A Few Good Agents (AFGA) that motivates self-interested agents to cooperate with one another to reduce the price of anarchy, while bounding the price of monarchy at the same time. A preliminary set of experiments on the El Farol bar problem, a simple example of MARS, show promising results.

Multi-agent learning is a crucial method to control or find solutions for systems, in which more than one entity needs to be adaptive. In today's interconnected world, such systems are ubiquitous in many domains, including auctions in economics, swarm robotics in computer science, and politics in social sciences. Multi-agent learning is inherently more complex than single-agent learning and has a relatively thin theoretical framework supporting it. Recently, multi-agent learning dynamics have been linked to evolutionary game theory, allowing the interpretation of learning as an evolution of competing policies in the mind of the learning agents. The dynamical system from evolutionary game theory that has been linked to Q-learning predicts the expected behavior of the learning agents. Closer analysis however allows for two interesting observations: the predicted behavior is not always the same as the actual behavior, and in case of deviation, the predicted behavior is more desirable. This discrepancy is elucidated in this article, and based on these new insights Frequency Adjusted Q- (FAQ-) learning is proposed. This variation of Q-learning perfectly adheres to the predictions of the evolutionary model for an arbitrarily large part of the policy space. In addition to the theoretical discussion, experiments in the three classes of two-agent two-action games illustrate the superiority of FAQ-learning.

Robots had a great impact on the manufacturing industry ever since the early seventies when companies such as KUKA and ABB started deploying their first industrial robots. These robots merely performed very specific tasks in specific ways within well-defined environments. Still, they proved to be very useful as they could exceed human performance at these tasks. However, in order to enable robots to enter our daily life, they need to become more versatile and need to operate in much less structured environments. This thesis is partly devoted to stretching these limitations by means of learning, namely imitation learning (IL) and inverse reinforcement learning (IRL). Reinforcement learning (RL) is a powerful approach to enable robots to solve a task in an unknown environment. The practitioner describes a desired behavior by specifying a reward function and the robot autonomously interacts with the environment in order to find a control policy that generates high accumulated reward. However, RL is not suitable for teaching new tasks by non-experts because specifying appropriate reward functions can be difficult. Demonstrating the desired behavior is often easier for non-experts. Imitation learning can be used in order to enable the robot to reproduce the demonstrations. However, without explicitly inferring and modeling the intentions of the demonstrations, it can become difficult to solve the task for unseen situations. Inverse reinforcement learning (IRL) therefore aims to infer a reward function from the demonstrations, such that optimizing this reward function yields the desired behavior even for different situations. This thesis introduces a unifying approach to solve the inverse reinforcement learning problem in the same way as the reinforcement learning problem. This is achieved by framing both problems as information projection problems, i.e., we strive to minimize the relative entropy between a probabilistic model of the robot behavior and a given desired distribution. Furthermore, a trust region on the robot behavior is used to stabilize the optimization. For inverse reinforcement learning, the desired distribution is implicitly given by the expert demonstrations. The resulting optimization can be efficiently solved using state-of-the-art reinforcement learning methods. For reinforcement learning, the log-likelihood of the desired distribution is given by the reward function. The resulting optimization problem corresponds to a standard reinforcement learning formulation, except for an additional objective of maximizing the entropy of the robot behavior. This entropy objective adds little overhead to the optimization, but can lead to better exploration and more diversified policies. Trust-region I-projections are not only useful for training robots, but can also be applied to other machine learning problems. I-projections are typically used for variational inference, in order to approximate an intractable distribution by a simpler model. However, the resulting optimization problems are usually optimized based on stochastic gradient descent which often suffers from high variance in the gradient estimates. As trust-region I-projections where shown to be effective for reinforcement learning and inverse reinforcement learning, this thesis also explores their use for variational inference. More specifically, trust-region I-projections are investigated for the problem of approximating an intractable distribution by a Gaussian mixture model (GMM) with an adaptive number of components. GMMs are highly desirable for variational inference because they can yield arbitrary accurate approximations while inference from GMMs is still relatively cheap. In order to make learning the GMM feasible, we derive a lower bound that enables us to decompose the objective function. The optimization can then be performed by iteratively updating individual components using a technique from reinforcement learning. The resulting method is capable of learning approximations of significantly higher quality than existing variational inference methods. Due to the similarity of the underlying optimization problems, the insights gained from our variational inference method are also useful for IL and IRL. Namely, a similar lower bound can be applied also for the I-projection formulation of imitation learning. However, whereas for variational inference the lower bound serves to decompose the objective function, for imitation learning it allows us to provide a reward signal to the robot that does not depend on its behavior. Compared to reward functions that are relative to the current behavior of the robot---which are typical for popular adversarial methods---behavior-independent reward functions have the advantages that we can show convergence even for greedy optimization. Furthermore, behavior-independent reward functions solve the inverse reinforcement learning problem, thereby closing the gap between imitation learning and IRL. However, algorithms derived from our non-adversarial formulation are actually very similar to existing AIL methods, and we can even show that adversarial inverse reinforcement learning (AIRL) is indeed an instance of our formulation. AIRL was derived from an adversarial formulation, and we point out several problems of that derivation. In contrast, we show that AIRL can be straightforwardly derived from out non-adversarial formulation. Furthermore, we demonstrate that the non-adversarial formulation can be also used to derive novel algorithms by presenting a non-adversarial method for offline imitation learning.

Although reinforcement learning (RL) has been successfully deployed in a variety of tasks, learning speed remains a fundamental problem for applying RL in complex environments. Transfer learning aims to ameliorate this shortcoming by speeding up learning through the adaptation of previously learned behaviors in similar tasks. Transfer techniques often use an inter-task mapping, which determines how a pair of tasks are related. Instead of relying on a hand-coded inter-task mapping, this paper proposes a novel transfer learning method capable of autonomously creating an inter-task mapping by using a novel combination of sparse coding, sparse projection learning and sparse Gaussian processes. We also propose two new transfer algorithms (TrLSPI and TrFQI) based on least squares policy iteration and fitted-Q-iteration. Experiments not only show successful transfer of information between similar tasks, inverted pendulum to cart pole, but also between two very different domains: mountain car to cart pole. This paper empirically shows that the learned inter-task mapping can be successfully used to (1) improve the performance of a learned policy on a fixed number of environmental samples, (2) reduce the learning times needed by the algorithms to converge to a policy on a fixed number of samples, and (3) converge faster to a near-optimal policy given a large number of samples.

Aim and ApproachLarge cooling water system has great potential of energy saving due to improper operation control. The coupled hydraulic and thermodynamic characteristics among chillers, pumps, and cooling towers lead to the difficulty of cooling water system operation. This study investigates an advanced control approach of multi-agent reinforcement learning (RL) for a large cooling water system. The multiple RL agents interact with each other to deal with the complex coupled characteristics in the cooling water system. The proposed approach is implemented in a real-world cooling water system and validated with experiments.Scientific Innovation and RelevanceIn this study, a multi-agent RL approach is proposed to separately control the chillers, pumps, and cooling towers in the cooling water system. The soft actor critic algorithm is used in the RL agent, as it is an efficient and advanced RL algorithm for continuous and discrete control problems. The agents are designed to have private states, private actions, and shared rewards. The reward functions are carefully designed to balance the agents for global optima.To provide a realistic environment for RL training, a detailed physical modeling platform of a cooling water system is established and validated. This platform is capable of hydraulic and thermodynamic calculation of cooling water systems with customized numbers and characteristics of chillers, pumps, and cooling towers, which is scalable for different cooling water systems.The proposed multi-agent RL controller is implemented in the building automation system of a real cooling water system. A one-month experimental study of the multi-agent RL performance is conducted. The proposed multi-agent RL controller is compared with a single-agent RL controller and a rule-based controller through real-world experiments to demonstrate the energy performance. This is a pilot study of multi-agent RL approach to real-world cooling water system control.Preliminary Results and ConclusionsThe preliminary simulation study is conducted in the established cooling water system model. The electricity consumption simulation error is 4.0% in coefficient of variation of mean absolute error (CVMAE) compared with the measured 15-minute electricity consumption. The 6-month simulation shows that the single-agent RL can save 6.2% energy from the rule-based control, while the multi-agent RL can save 7.2% electricity consumption of the cooling water system. In the following study, the multi-agent RL implementation will be practiced with more experiments and applications for cooling water system control.

Reinforcement learning (RL) and multi-agent reinforcement learning (MARL) are disciplines concerned with defining automatically the behaviour of an agent, or a set of interacting agents, by means of reward mechanisms coming from the environment. An important research issue in the context of RL and MARL is the definition of approaches to combine the knowledge of multiple learning agents to improve the overall performance of the multi-agent system (MAS). This paper illustrates how to improve RL and MARL algorithms by utilizing results from multi-linear algebra such as tensors and tensor factorizations. In particular, the focus is on showing how to modify existing algorithms from literature to include a tensor factorization step applied to the Q-Tables learned by the individual agents to generalize the knowledge about the actions performed in the environment. The modified algorithms are then evaluated in three RL and MARL scenarios against their unmodified version to show the benefits of the tensor factorization step.

Reinforcement learning (RL) is an area of research that has blossomed tremendously in recent years and has shown remarkable potential for artificial intelligence based opponents in computer games. This success is primarily due to the vast capabilities of convolutional neural networks, that can extract useful features from noisy and complex data. Games are excellent tools to test and push the boundaries of novel RL algorithms because they give valuable insight into how well an algorithm can perform in isolated environments without the real-life consequences. Real-time strategy games (RTS) is a genre that has tremendous complexity and challenges the player in short and long-term planning. There is much research that focuses on applied RL in RTS games, and novel advances are therefore anticipated in the not too distant future. However, there are to date few environments for testing RTS AIs. Environments in the literature are often either overly simplistic, such as microRTS, or complex and without the possibility for accelerated learning on consumer hardware like StarCraft II. This paper introduces the Deep RTS game environment for testing cutting-edge artificial intelligence algorithms for RTS games. Deep RTS is a high-performance RTS game made specifically for artificial intelligence research. It supports accelerated learning, meaning that it can learn at a magnitude of 50 000 times faster compared to existing RTS games. Deep RTS has a flexible configuration, enabling research in several different RTS scenarios, including partially observable state-spaces and map complexity. We show that Deep RTS lives up to our promises by comparing its performance with microRTS, ELF, and StarCraft II on high-end consumer hardware. Using Deep RTS, we show that a Deep Q-Network agent beats random-play agents over 70% of the time. Deep RTS is publicly available at https://github.com/cair/DeepRTS.