In recent years, multi-agent deep reinforcement learning has been successfully applied to various complicated scenarios such as computer games and robot swarms. We thoroughly study and compare the state-of-the-art cooperative multi-agent deep reinforcement learning algorithms. Specifically, we investigate the consequences of the "hyperparameter tricks" of QMIX and its improved variants. Our results show that: (1) The significant performance improvements of these variant algorithms come from hyperparameter-level optimizations in their open-source codes (2) After modest tuning and with no changes to the network architecture, QMIX can attain extraordinarily high win rates in all hard and super hard scenarios of StarCraft Multi-Agent Challenge (SMAC) and achieve state-of-the-art (SOTA). In this work, we proposed a reliable QMIX benchmark, which will be of great benefit to subsequent research. Besides, we proposed a hypothesis to explain the excellent performance of QMIX.

With Artificial Intelligence on the rise, human interaction with autonomous agents becomes more frequent. Effective human-agent collaboration requires that the human understands the agent's behavior, as failing to do so may lead to reduced productiveness, misuse, frustration and even danger. Agent strategy summarization methods are used to describe the strategy of an agent to its destined user through demonstration. The summary's purpose is to maximize the user's understanding of the agent's aptitude by showcasing its behaviour in a set of world states, chosen by some importance criteria. While shown to be useful, we show that these methods are limited in supporting the task of comparing agent behavior, as they independently generate a summary for each agent. In this paper, we propose a novel method for generating contrastive summaries that highlight the differences between agent's policies by identifying and ranking states in which the agents disagree on the best course of action. We conduct a user study in which participants face an agent selection task. Our results show that the novel disagreement-based summaries lead to improved user performance compared to summaries generated using HIGHLIGHTS, a previous strategy summarization algorithm.

Balancing safety and efficiency when planning in dense traffic is challenging. Interactive behavior planners incorporate prediction uncertainty and interactivity inherent to these traffic situations. Yet, their use of single-objective optimality impedes interpretability of the resulting safety goal. Safety envelopes which restrict the allowed planning region yield interpretable safety under the presence of behavior uncertainty, yet, they sacrifice efficiency in dense traffic due to conservative driving. Studies show that humans balance safety and efficiency in dense traffic by accepting a probabilistic risk of violating the safety envelope. In this work, we adopt this safety objective for interactive planning. Specifically, we formalize this safety objective, present the Risk-Constrained Robust Stochastic Bayesian Game modeling interactive decisions satisfying a maximum risk of violating a safety envelope under uncertainty of other traffic participants' behavior and solve it using our variant of Multi-Agent Monte Carlo Tree Search. We demonstrate in simulation that our approach outperforms baselines approaches, and by reaching the specified violation risk level over driven simulation time, provides an interpretable and tunable safety objective for interactive planning.

In this paper, we study the problem of deceptive reinforcement learning to preserve the privacy of a reward function. Reinforcement learning is the problem of finding a behaviour policy based on rewards received from exploratory behaviour. A key ingredient in reinforcement learning is a reward function, which determines how much reward (negative or positive) is given and when. However, in some situations, we may want to keep a reward function private; that is, to make it difficult for an observer to determine the reward function used. We define the problem of privacy-preserving reinforcement learning, and present two models for solving it. These models are based on dissimulation -- a form of deception that `hides the truth'. We evaluate our models both computationally and via human behavioural experiments. Results show that the resulting policies are indeed deceptive, and that participants can determine the true reward function less reliably than that of an honest agent.

There has been rapidly growing interest in meta-learning as a method for increasing the flexibility and sample efficiency of reinforcement learning. One problem in this area of research, however, has been a scarcity of adequate benchmark tasks. In general, the structure underlying past benchmarks has either been too simple to be inherently interesting, or too ill-defined to support principled analysis. In the present work, we introduce a new benchmark for meta-RL research, which combines structural richness with structural transparency. Alchemy is a 3D video game, implemented in Unity, which involves a latent causal structure that is resampled procedurally from episode to episode, affording structure learning, online inference, hypothesis testing and action sequencing based on abstract domain knowledge. We evaluate a pair of powerful RL agents on Alchemy and present an in-depth analysis of one of these agents. Results clearly indicate a frank and specific failure of meta-learning, providing validation for Alchemy as a challenging benchmark for meta-RL. Concurrent with this report, we are releasing Alchemy as public resource, together with a suite of analysis tools and sample agent trajectories.

The expansion of artificial intelligence (AI) and autonomous systems has shown the potential to generate enormous social good while also raising serious ethical and safety concerns. AI technology is increasingly adopted in transportation. A survey of various in-vehicle technologies found that approximately 64% of the respondents used a smartphone application to assist with their travel. The top-used applications were navigation and real-time traffic information systems. Among those who used smartphones during their commutes, the top-used applications were navigation and entertainment. There is a pressing need to address relevant social concerns to allow for the development of systems of intelligent agents that are informed and cognizant of ethical standards. Doing so will facilitate the responsible integration of these systems in society. To this end, we have applied Multi-Criteria Decision Analysis (MCDA) to develop a formal Multi-Attribute Impact Assessment (MAIA) questionnaire for examining the social and ethical issues associated with the uptake of AI. We have focused on the domain of autonomous vehicles (AVs) because of their imminent expansion. However, AVs could serve as a stand-in for any domain where intelligent, autonomous agents interact with humans, either on an individual level (e.g., pedestrians, passengers) or a societal level.

We introduce a novel method to aggregate Bipolar Argumentation (BA) Frameworks expressing opinions by different parties in debates. We use Bipolar Assumption-based Argumentation (ABA) as an all-encompassing formalism for BA under different semantics. By leveraging on recent results on judgement aggregation in Social Choice Theory, we prove several preservation results, both positive and negative, for relevant properties of Bipolar ABA.

Egalitarian considerations play a central role in many areas of social choice theory. Applications of egalitarian principles range from ensuring everyone gets an equal share of a cake when deciding how to divide it, to guaranteeing balance with respect to gender or ethnicity in committee elections. Yet, the egalitarian approach has received little attention in judgment aggregation -- a powerful framework for aggregating logically interconnected issues. We make the first steps towards filling that gap. We introduce axioms capturing two classical interpretations of egalitarianism in judgment aggregation and situate these within the context of existing axioms in the pertinent framework of belief merging. We then explore the relationship between these axioms and several notions of strategyproofness from social choice theory at large. Finally, a novel egalitarian judgment aggregation rule stems from our analysis; we present complexity results concerning both outcome determination and strategic manipulation for that rule.

In multi-agent reinforcement learning, the problem of learning to act is particularly difficult because the policies of co-players may be heavily conditioned on information only observed by them. On the other hand, humans readily form beliefs about the knowledge possessed by their peers and leverage beliefs to inform decision-making. Such abilities underlie individual success in a wide range of Markov games, from bluffing in Poker to conditional cooperation in the Prisoner's Dilemma, to convention-building in Bridge. Classical methods are usually not applicable to complex domains due to the intractable nature of hierarchical beliefs (i.e. beliefs of other agents' beliefs). We propose a scalable method to approximate these belief structures using recursive deep generative models, and to use the belief models to obtain representations useful to acting in complex tasks. Our agents trained with belief models outperform model-free baselines with equivalent representational capacity using common training paradigms. We also show that higher-order belief models outperform agents with lower-order models.

Can artificial agents benefit from human conventions? Human societies manage to successfully self-organize and resolve the tragedy of the commons in common-pool resources, in spite of the bleak prediction of non-cooperative game theory. On top of that, real-world problems are inherently large-scale and of low observability. One key concept that facilitates human coordination in such settings is the use of conventions. Inspired by human behavior, we investigate the learning dynamics and emergence of temporal conventions, focusing on common-pool resources. Extra emphasis was given in designing a realistic evaluation setting: (a) environment dynamics are modeled on real-world fisheries, (b) we assume decentralized learning, where agents can observe only their own history, and (c) we run large-scale simulations (up to 64 agents). Uncoupled policies and low observability make cooperation hard to achieve; as the number of agents grow, the probability of taking a correct gradient direction decreases exponentially. By introducing an arbitrary common signal (e.g., date, time, or any periodic set of numbers) as a means to couple the learning process, we show that temporal conventions can emerge and agents reach sustainable harvesting strategies. The introduction of the signal consistently improves the social welfare (by 258% on average, up to 3306%), the range of environmental parameters where sustainability can be achieved (by 46% on average, up to 300%), and the convergence speed in low abundance settings (by 13% on average, up to 53%).