Multi-player mean-payoff games are a natural formalism for modelling the behaviour of concurrent and multi-agent systems with self-interested players. Players in such a game traverse a graph, while attempting to maximise a (mean-)payoff function that depends on the play generated. As with all games, the equilibria that could arise may have undesirable properties. However, as system designers, we typically wish to ensure that equilibria in such systems correspond to desirable system behaviours, for example, satisfying certain safety or liveness properties. One natural way to do this would be to specify such desirable properties using temporal logic. Unfortunately, the use of temporal logic specifications causes game theoretic verification problems to have very high computational complexity. To address this issue, we consider ω-regular specifications. These offer a concise and intuitive way of specifying system behaviours with a comparatively low computational overhead. The main results of this work are characterisation and complexity bounds for the problem of determining if there are equilibria that satisfy a given ω-regular specification in a multi-player mean-payoff game in a number of computationally relevant game-theoretic settings.
ABSTRACTA robust testing program is necessary for containing the spread of COVID-19 infections before a vaccine becomes available. However, due to an acute shortage of testing kits (especially in low-resource developing countries), designing an optimal testing program/strategy is a challenging problem to solve. Prior literature on testing strategies suffers from two major limitations: (i) it does not account for the trade-off between testing of symptomatic and asymptomatic individuals, and (ii) it primarily focuses on static testing strategies, which leads to significant shortcomings in the testing program’s effectiveness. In this paper, we address these limitations by making five novel contributions. (i) We formally define the optimal testing problem and propose the DOCTOR POMDP model to tackle it. (ii) We solve the DOCTOR POMDP using a scalable Monte Carlo tree search based algorithm. (iii) We provide a rigorous experimental analysis of DOCTOR’s testing strategies against static baselines - our results show that when applied to the city of Santiago in Panama, DOCTOR’s strategies result in ∼40% fewer COVID-19 infections (over one month) as compared to state-of-the-art static baselines. (iv) In addition, we analyze DOCTOR’s testing policy to derive insights about the reasons behind the optimality of DOCTOR’s testing policy. (v) Finally, we characterize conditions (of the real world) under which DOCTOR’s optimization would be of most benefit to government policy makers, and thus requires significant attention from researchers in this area. Our work complements the growing body of research on COVID-19, and serves as a proof-of-concept that illustrates the benefit of having an AI-driven adaptive testing strategy for COVID-19.
Deploying reinforcement learning (RL) involves major concerns around safety. Engineering a reward signal that allows the agent to maximize its performance while remaining safe is not trivial. Safe RL studies how to mitigate such problems. For instance, we can decouple safety from reward using constrained Markov decision processes (CMDPs), where an independent signal models the safety aspects. In this setting, an RL agent can autonomously find tradeoffs between performance and safety. Unfortunately, most RL agents designed for CMDPs only guarantee safety after the learning phase, which might prevent their direct deployment. In this work, we investigate settings where a concise abstract model of the safety aspects is given, a reasonable assumption since a thorough under-standing of safety-related matters is a prerequisite for deploying RL in typical applications. Factored CMDPs provide such compact models when a small subset of features describe the dynamics relevant for the safety constraints. We propose an RL algorithm that uses this abstract model to learn policies for CMDPs safely, that is without violating the constraints. During the training process, this algorithm can seamlessly switch from a conservative policy to a greedy policy without violating the safety constraints. We prove that this algorithm is safe under the given assumptions. Empirically, we show that even if safety and reward signals are contradictory, this algorithm always operates safely and, when they are aligned, this approach also improves the agent’s performance.
In service robot applications, planning is often integrated with task allocation. Linear Temporal Logic (LTL) as an expressive high-level formalism is widely used for task specification, and allows for formalised restrictions on temporal sequences of tasks. In multiagent planning, a Multi-Objective Markov Decision Process extends the standard model with vector rewards capturing possibly conflicting planning objectives. Such objectives include the success rates of accomplishing individual tasks, and the cost budgets for individual agents. In this paper, we consider the problem of concurrently allocating LTL task sequences to a team of agents and calculating optimal task schedulers simultaneously, satisfying cost and probability thresholds. We reduce this problem to multi-objective scheduler synthesis for a team MDP structure, whose size is linear in the number of agents. Our preliminary experiment demonstrates the scalability of our approach.
This paper addresses an epidemiologic inference problem where, given realtime observation of test results, presence of symptoms, and physical contacts, the most likely infected individuals need to be inferred. The inference problem is modeled as a hidden Markov model where infection probabilities are updated at every time step and evolve between time steps. We suggest a unique inference approach that avoids storing the given observations explicitly. Theoretical justification for the proposed model is provided under specific simplifying assumptions. To complement these theoretical results, a comprehensive experimental study is performed using a custom-built agent-based simulator that models inter-agent contacts. The reported results show the effectiveness of the proposed inference model when considering more realistic scenarios – where the simplifying assumptions do not hold. When pairing the proposed inference model with a simple testing and quarantine policy, promising trends are obtained where the epidemic progression is significantly slowed down while quarantining a bounded number of individuals. © 2021 International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org). All rights reserved.
Relational agents who personalise their message to individual users need to learn about the user first. In the context of an agent who personalises the inclusion of 10 relational cues in its dialogue, we explore training the agent based on first time user’s responses to a single example of each cue prior to the session with the virtual advisor. We designed a between subjects study with three groups: Empathic (all relational cues); Neutral (no relational cues) and Adaptive (only helpful cues included). We found that in the Adaptive group, students received what they found helpful more often than in the other two groups. Analysis of the discrepancy between what user’s found helpful and what they received was least in the Adaptive group and greatest in the Neutral group.
Goal recognition is the problem of determining an agent’s intent by observing its actions. In the context of AI research, the problem is tackled for two quite different purposes: to determine an agent’s most probable goal or, for human-aware planning including planned—or strategic—deception, to determine an observer’s most likely belief about that goal. Making no distinction, contemporary models tend to assume an infallible observer, deceived only while it has limited access to information or if the environment itself is only partially observable. Focusing on the second purpose, we propose an extended framework that incorporates formal definitions of confirmation bias, selective attention and memory decay. In contrast to pre-existing models, our approach combines explicit consideration of prior probabilities with a principled representation of observer confidence and distinguishes between potential observations—i.e., every observable event within the observer’s frame of reference—and recalled observations which we model as a function of attention and memory. We show that when these factors are taken into consideration, false beliefs may arise and can be made to persist, even in a fully observable environment—thus providing a perceptual model readily incorporated into the “thinking” of an adversarial agent for the purpose of strategic deception.
The pursuit of values drives human behavior and promotes cooperation. Existing research is focused on general (e.g., Schwartz) values that transcend contexts. However, context-specific values are necessary to (1) understand human decisions, and (2) engineer intelligent agents that can elicit human values and take value-aligned actions. We propose Axies, a hybrid (human and AI) methodology to identify context-specific values. Axies simplifies the abstract task of value identification as a guided value annotation process involving human annotators. Axies exploits the growing availability of valueladen text corpora and Natural Language Processing to assist the annotators in systematically identifying context-specific values. We evaluate Axies in a user study involving 60 subjects. In our study, six annotators generate value lists for two timely and important contexts: Covid-19 measures, and sustainable Energy. Then, two policy experts and 52 crowd workers evaluate Axies value lists. We find that Axies yields values that are context-specific, consistent across different annotators, and comprehensible to end users.
In safe Reinforcement Learning (RL), the agent attempts to find policies which maximize the expectation of accumulated rewards and guarantee its safety to remain above a given threshold. Hence, it is straightforward to formalize safe RL problems by both a reward function and a safety constraint. We define safety as the probability of survival in environments where taking risky actions could lead to early termination of the task. Although the optimization problem is already constrained by a safety threshold, reward signals related to unsafe terminal states influence the original maximization objective of the task. Selecting the appropriate value of these signals is often a time consuming and challenging reward engineering task, which requires expert knowledge of the domain. This paper presents a safe RL algorithm, called Masked Constrained Policy Optimization (MCPO), in which the learning process is constrained by safety and excludes the risk reward signals. We develop MCPO as an extension of gradient-based policy search methods, in which the updates of the policy and the expected reward models are masked. Our method benefits from having a high probability of satisfying the given constraints for every policy in the learning process. We validate the proposed algorithm in two continuous tasks. Our findings prove the proposed algorithm is able to neglect risk reward signals, and thereby resolving the desired safety-performance trade-off without having the need for hand-tuning rewards.
In swarm robotics, research mainly follows a theoretical approach that considers robot systems in the abstract, where the complexity and capabilities of the underlying model are often reduced to their minimum. One general and well-investigated model is øblot, where the robots are silent, anonymous, and oblivious. In this work, we introduce MOBLOT, a model that extends øblot to address a larger spectrum of cases. MOBLOT stands for molecular oblivious robots: like atoms combine themselves to form molecules, in MOBLOT simple robots can move to form more complex computational units, having an extent and different capabilities with respect to robots; like molecules combine themselves to form the matter, in MOBLOT the complex structures can exploit their own capabilities to arrange themselves to form any shape defining an acceptable final structure. In MOBLOT, we formally define the Matter Formation (MF) problem and, as a preliminary general result, we provide a necessary condition for its solvability which relies on symmetricity. Informally, the symmetricity of a configuration measures the amount of symmetries of the robots' disposal. We actually show how dealing with molecules can resolve in some cases the symmetry breaking issue where øblot cannot. Finally, we provide a case study for MOBLOT, that is, a representative MF problem along with a resolution distributed algorithm.
We present an extension of the organizational model and infrastructure adopted in JaCaMo, that explicitly encompasses the notion of exception. We propose an exception handling mechanism for organization management in multi-agent systems. This mechanism relies on abstractions that are seamlessly integrated with organizational concepts, such as responsibilities, goals and norms.
Embodied conversational agents (ECAs) benefit from non-verbal behavior for natural and efficient interaction with users. Gesticulation - hand and arm movements accompanying speech - is an essential part of non-verbal behavior. Gesture generation models have been developed for several decades: starting with rule-based and ending with mainly data-driven methods. To date, recent end-to-end gesture generation methods have not been evaluated in a real-time interaction with users. We present a proof-of-concept framework, which is intended to facilitate evaluation of modern gesture generation models in interaction. We demonstrate an extensible open-source framework that contains three components: 1) a 3D interactive agent; 2) a chatbot backend; 3) a gesticulating system. Each component can be replaced, making the proposed framework applicable for investigating the effect of different gesturing models in real-time interactions with different communication modalities, chatbot backends, or different agent appearances. The code and video are available at the project page https://nagyrajmund.github.io/project/gesturebot.
Many real-world scenarios involve teams of agents that have to coordinate their actions to reach a shared goal. We focus on the setting in which a team of agents faces an opponent in a zero-sum, imperfect-information game. Team members can coordinate their strategies before the beginning of the game, but are unable to communicate during the playing phase of the game. This is the case, for example, in Bridge, collusion in poker, and collusion in bidding. In this setting, model-free RL methods are oftentimes unable to capture coordination because agents' policies are executed in a decentralized fashion. Our first contribution is a game-theoretic centralized training regimen to effectively perform trajectory sampling so as to foster team coordination. When team members can observe each other actions, we show that this approach provably yields equilibrium strategies. Then, we introduce a signaling-based framework to represent team coordinated strategies given a buffer of past experiences. Each team member's policy is parametrized as a neural network whose output is conditioned on a suitable exogenous signal, drawn from a learned probability distribution. By combining these two elements, we empirically show convergence to coordinated equilibria in cases where previous state-of-the-art multi-agent RL algorithms did not.
Continual learning aims to provide intelligent agents capable of learning multiple tasks sequentially with neural networks. One of its main challenging, catastrophic forgetting, is caused by the neural networks non-optimal ability to learn in non-stationary distributions. In most settings of the current approaches, the agent starts from randomly initialized parameters and is optimized to master the current task regardless of the usefulness of the learned representation for future tasks. Moreover, each of the future tasks uses all the previously learned knowledge although parts of this knowledge might not be helpful for its learning. These cause interference among tasks, especially when the data of previous tasks is not accessible. In this paper, we propose a new method, named Self-Attention Meta-Learner (SAM), which learns a prior knowledge for continual learning that permits learning a sequence of tasks, while avoiding catastrophic forgetting. SAM incorporates an attention mechanism that learns to select the particular relevant representation for each future task. Each task builds a specific representation branch on top of the selected knowledge, avoiding the interference between tasks. We evaluate the proposed method on the Split CIFAR-10/100 and Split MNIST benchmarks in the task agnostic inference. We empirically show that we can achieve a better performance than several state-of-the-art methods for continual learning by building on the top of selected representation learned by SAM. We also show the role of the meta-attention mechanism in boosting informative features corresponding to the input data and identifying the correct target in the task agnostic inference. Finally, we demonstrate that popular existing continual learning methods gain a performance boost when they adopt SAM as a starting point.
In order for agents trained by deep reinforcement learning to work alongside humans in realistic settings, we will need to ensure that the agents are \emph{robust}. Since the real world is very diverse, and human behavior often changes in response to agent deployment, the agent will likely encounter novel situations that have never been seen during training. This results in an evaluation challenge: if we cannot rely on the average training or validation reward as a metric, then how can we effectively evaluate robustness? We take inspiration from the practice of \emph{unit testing} in software engineering. Specifically, we suggest that when designing AI agents that collaborate with humans, designers should search for potential edge cases in \emph{possible partner behavior} and \emph{possible states encountered}, and write tests which check that the behavior of the agent in these edge cases is reasonable. We apply this methodology to build a suite of unit tests for the Overcooked-AI environment, and use this test suite to evaluate three proposals for improving robustness. We find that the test suite provides significant insight into the effects of these proposals that were generally not revealed by looking solely at the average validation reward.
Agents that interact with other agents often do not know a priori what the other agents' strategies are, but have to maximise their own online return while interacting with and learning about others. The optimal adaptive behaviour under uncertainty over the other agents' strategies w.r.t. some prior can in principle be computed using the Interactive Bayesian Reinforcement Learning framework. Unfortunately, doing so is intractable in most settings, and existing approximation methods are restricted to small tasks. To overcome this, we propose to meta-learn approximate belief inference and Bayes-optimal behaviour for a given prior. To model beliefs over other agents, we combine sequential and hierarchical Variational Auto-Encoders, and meta-train this inference model alongside the policy. We show empirically that our approach outperforms existing methods that use a model-free approach, sample from the approximate posterior, maintain memory-free models of others, or do not fully utilise the known structure of the environment.
Precisely when the success of artificial intelligence (AI) sub-symbolic techniques makes them be identified with the whole AI by many non-computer-scientists and non-technical media, symbolic approaches are getting more and more attention as those that could make AI amenable to human understanding. Given the recurring cycles in the AI history, we expect that a revamp of technologies often tagged as “classical AI”—in particular, logic-based ones—will take place in the next few years. On the other hand, agents and multi-agent systems (MAS) have been at the core of the design of intelligent systems since their very beginning, and their long-term connection with logic-based technologies , which characterised their early days, might open new ways to engineer explainable intelligent systems . This is why understanding the current status of logic-based technologies for MAS is nowadays of paramount importance. Accordingly, this paper aims at providing a comprehensive view of those technologies by making them the subject of a systematic literature review (SLR). The resulting technologies are discussed and evaluated from two different perspectives: the MAS and the logic-based ones.
Partially Observable Monte-Carlo Planning (POMCP) is a powerful online algorithm able to generate approximate policies for large Partially Observable Markov Decision Processes. The online nature of this method supports scalability by avoiding complete policy representation. The lack of an explicit representation however hinders interpretability. In this work, we propose a methodology based on Satisfiability Modulo Theory (SMT) for analyzing POMCP policies by inspecting their traces, namely sequences of belief-action-observation triplets generated by the algorithm. The proposed method explores local properties of policy behavior to identify unexpected decisions. We propose an iterative process of trace analysis consisting of three main steps, i) the definition of a question by means of a parametric logical formula describing (probabilistic) relationships between beliefs and actions, ii) the generation of an answer by computing the parameters of the logical formula that maximize the number of satisfied clauses (solving a MAX-SMT problem), iii) the analysis of the generated logical formula and the related decision boundaries for identifying unexpected decisions made by POMCP with respect to the original question. We evaluate our approach on Tiger, a standard benchmark for POMDPs, and a real-world problem related to mobile robot navigation. Results show that the approach can exploit human knowledge on the domain, outperforming state-of-the-art anomaly detection methods in identifying unexpected decisions. An improvement of the Area Under Curve up to 47% has been achieved in our tests.
Policy gradient methods have become one of the most popular classes of algorithms for multi-agent reinforcement learning. A key challenge, however, that is not addressed by many of these methods is multi-agent credit assignment: assessing an agent’s contribution to the overall performance, which is crucial for learning good policies. We propose a novel algorithm called Dr.Reinforce that explicitly tackles this by combining difference rewards with policy gradients to allow for learning decentralized policies when the reward function is known. By differencing the reward function directly, Dr.Reinforce avoids difficulties associated with learning the Q -function as done by counterfactual multi-agent policy gradients (COMA), a state-of-the-art difference rewards method. For applications where the reward function is unknown, we show the effectiveness of a version of Dr.Reinforce that learns an additional reward network that is used to estimate the difference rewards.
We consider exploration tasks in which an autonomous mobile robot incrementally builds maps of initially unknown indoor environments. In such tasks, the robot makes a sequence of decisions on where to move next that, usually, are based on knowledge about the observed parts of the environment. In this paper, we present an approach that exploits a prediction of the geometric structure of the unknown parts of an environment to improve exploration performance. In particular, we leverage an existing method that reconstructs the layout of an environment starting from a partial grid map and that predicts the shape of partially observed rooms on the basis of geometric features representing the regularities of the indoor environment. Then, we originally employ the predicted layout to estimate the amount of new area the robot would observe from candidate locations in order to inform the selection of the next best location and to early stop the exploration when no further relevant area is expected to be discovered. Experimental activities show that our approach is able to exploit the predicted layout of partially observed rooms in order to speed up the exploration.
