The transfer of knowledge from one policy to another is an important tool in Deep Reinforcement Learning. This process, referred to as distillation, has been used to great success, for example, by enhancing the optimisation of agents, leading to stronger performance faster, on harder domains [26, 32, 5, 8]. Despite the widespread use and conceptual simplicity of distillation, many different formulations are used in practice, and the subtle variations between them can often drastically change the performance and the resulting objective that is being optimised. In this work, we rigorously explore the entire landscape of policy distillation, comparing the motivations and strengths of each variant through theoretical and empirical analysis. Our results point to three distillation techniques, that are preferred depending on specifics of the task. Specifically a newly proposed expected entropy regularised distillation allows for quicker learning in a wide range of situations, while still guaranteeing convergence.

Reinforcement learning (RL) algorithms have been around for decades and employed to solve various sequential decision-making problems. These algorithms however have faced great challenges when dealing with high-dimensional environments. The recent development of deep learning has enabled RL methods to drive optimal policies for sophisticated and capable agents, which can perform efficiently in these challenging environments. This paper addresses an important aspect of deep RL related to situations that require multiple agents to communicate and cooperate to solve complex tasks. A survey of different approaches to problems related to multi-agent deep RL (MADRL) is presented, including non-stationarity, partial observability, continuous state and action spaces, multi-agent training schemes, multi-agent transfer learning. The merits and demerits of the reviewed methods will be analyzed and discussed, with their corresponding applications explored. It is envisaged that this review provides insights about various MADRL methods and can lead to future development of more robust and highly useful multi-agent learning methods for solving real-world problems.

In this paper, we argue that simulation platforms enable a novel type of embodied spatial reasoning, one facilitated by a formal model of object and event semantics that renders the continuous quantitative search space of an open-world, real-time environment tractable. We provide examples for how a semantically-informed AI system can exploit the precise, numerical information provided by a game engine to perform qualitative reasoning about objects and events, facilitate learning novel concepts from data, and communicate with a human to improve its models and demonstrate its understanding. We argue that simulation environments, and game engines in particular, bring together many different notions of "simulation" and many different technologies to provide a highly-effective platform for developing both AI systems and tools to experiment in both machine and human intelligence.

Dungeon Crawl Stone Soup is a popular, single-player, free and open-source rogue-like video game with a sufficiently complex decision space that makes it an ideal testbed for research in cognitive systems and, more generally, artificial intelligence. This paper describes the properties of Dungeon Crawl Stone Soup that are conducive to evaluating new approaches of AI systems. We also highlight an ongoing effort to build an API for AI researchers in the spirit of recent game APIs such as MALMO, ELF, and the Starcraft II API. Dungeon Crawl Stone Soup's complexity offers significant opportunities for evaluating AI and cognitive systems, including human user studies. In this paper we provide (1) a description of the state space of Dungeon Crawl Stone Soup, (2) a description of the components for our API, and (3) the potential benefits of evaluating AI agents in the Dungeon Crawl Stone Soup video game.

This is an integrative review that address the question, "What makes for a good explanation?" with reference to AI systems. Pertinent literatures are vast. Thus, this review is necessarily selective. That said, most of the key concepts and issues are expressed in this Report. The Report encapsulates the history of computer science efforts to create systems that explain and instruct (intelligent tutoring systems and expert systems). The Report expresses the explainability issues and challenges in modern AI, and presents capsule views of the leading psychological theories of explanation. Certain articles stand out by virtue of their particular relevance to XAI, and their methods, results, and key points are highlighted. It is recommended that AI/XAI researchers be encouraged to include in their research reports fuller details on their empirical or experimental methods, in the fashion of experimental psychology research reports: details on Participants, Instructions, Procedures, Tasks, Dependent Variables (operational definitions of the measures and metrics), Independent Variables (conditions), and Control Conditions.

Despite the notable successes in video games such as Atari 2600, current AI is yet to defeat human champions in the domain of real-time strategy (RTS) games. One of the reasons is that an RTS game is a multi-agent game, in which single-agent reinforcement learning methods cannot simply be applied because the environment is not a stationary Markov Decision Process. In this paper, we present a first step toward finding a game-theoretic solution to RTS games by applying Neural Fictitious Self-Play (NFSP), a game-theoretic approach for finding Nash equilibria, to Mini-RTS, a small but nontrivial RTS game provided on the ELF platform. More specifically, we show that NFSP can be effectively combined with policy gradient reinforcement learning and be applied to Mini-RTS. Experimental results also show that the scalability of NFSP can be substantially improved by pretraining the models with simple self-play using policy gradients, which by itself gives a strong strategy despite its lack of theoretical guarantee of convergence.

In this paper, we propose a framework that enables a human teacher to shape a robot behaviour by interactively providing it with unlabeled instructions. We ground the meaning of instruction signals in the task learning process, and use them simultaneously for guiding the latter. We implement our framework as a modular architecture, named TICS (Task-Instruction-Contingency-Shaping) that combines different information sources: a predefined reward function, human evaluative feedback and unlabeled instructions. This approach provides a novel perspective for robotic task learning that lies between Reinforcement Learning and Supervised Learning paradigms. We evaluate our framework both in simulation and with a real robot. The experimental results demonstrate the effectiveness of our framework in accelerating the task learning process and in reducing the amount of required teaching signals.

Many real-world reinforcement learning tasks require multiple agents to make sequential decisions under the agents' interaction, where well-coordinated actions among the agents are crucial to achieve the target goal better at these tasks. One way to accelerate the coordination effect is to enable multiple agents to communicate with each other in a distributed manner and behave as a group. In this paper, we study a practical scenario when (i) the communication bandwidth is limited and (ii) the agents share the communication medium so that only a restricted number of agents are able to simultaneously use the medium, as in the state-of-the-art wireless networking standards. This calls for a certain form of communication scheduling. In that regard, we propose a multi-agent deep reinforcement learning framework, called SchedNet, in which agents learn how to schedule themselves, how to encode the messages, and how to select actions based on received messages. SchedNet is capable of deciding which agents should be entitled to broadcasting their (encoded) messages, by learning the importance of each agent's partially observed information. We evaluate SchedNet against multiple baselines under two different applications, namely, cooperative communication and navigation, and predator-prey. Our experiments show a non-negligible performance gap between SchedNet and other mechanisms such as the ones without communication and with vanilla scheduling methods, e.g., round robin, ranging from 32% to 43%.

In this paper, we take a first step towards generalizing a recently proposed method for dealing with the problem of convergence to incorrect equilibrium points of distance-based formation controllers. Specifically, we introduce a distance and area-based scheme for the formation control of $n$-agent systems in two dimensions using directed graphs and the single-integrator model. We show that under certain conditions on the edge lengths of the triangulated desired formation, the control ensures almost-global convergence to the correct formation.

Recent efforts on training visual navigation agents conditioned on language using deep reinforcement learning have been successful in learning policies for different multimodal tasks, such as semantic goal navigation and embodied question answering. In this paper, we propose a multitask model capable of jointly learning these multimodal tasks, and transferring knowledge of words and their grounding in visual objects across the tasks. The proposed model uses a novel Dual-Attention unit to disentangle the knowledge of words in the textual representations and visual concepts in the visual representations, and align them with each other. This disentangled task-invariant alignment of representations facilitates grounding and knowledge transfer across both tasks. We show that the proposed model outperforms a range of baselines on both tasks in simulated 3D environments. We also show that this disentanglement of representations makes our model modular, interpretable, and allows for transfer to instructions containing new words by leveraging object detectors.