Modern reinforcement learning algorithms can learn solutions to increasingly difficult control problems while at the same time reduce the amount of prior knowledge needed for their application. One of the remaining challenges is the definition of reward schemes that appropriately facilitate exploration without biasing the solution in undesirable ways, and that can be implemented on real robotic systems without expensive instrumentation. In this paper we focus on a setting in which goal tasks are defined via simple sparse rewards, and exploration is facilitated via agent-internal auxiliary tasks. We introduce the idea of simple sensor intentions (SSIs) as a generic way to define auxiliary tasks. SSIs reduce the amount of prior knowledge that is required to define suitable rewards. They can further be computed directly from raw sensor streams and thus do not require expensive and possibly brittle state estimation on real systems. We demonstrate that a learning system based on these rewards can solve complex robotic tasks in simulation and in real world settings. In particular, we show that a real robotic arm can learn to grasp and lift and solve a Ball-in-a-Cup task from scratch, when only raw sensor streams are used for both controller input and in the auxiliary reward definition.

Traditional distributed deep reinforcement learning (RL) commonly relies on exchanging the experience replay memory (RM) of each agent. Since the RM contains all state observations and action policy history, it may incur huge communication overhead while violating the privacy of each agent. Alternatively, this article presents a communication-efficient and privacy-preserving distributed RL framework, coined federated reinforcement distillation (FRD). In FRD, each agent exchanges its proxy experience replay memory (ProxRM), in which policies are locally averaged with respect to proxy states clustering actual states. To provide FRD design insights, we present ablation studies on the impact of ProxRM structures, neural network architectures, and communication intervals. Furthermore, we propose an improved version of FRD, coined mixup augmented FRD (MixFRD), in which ProxRM is interpolated using the mixup data augmentation algorithm. Simulations in a Cartpole environment validate the effectiveness of MixFRD in reducing the variance of mission completion time and communication cost, compared to the benchmark schemes, vanilla FRD, federated reinforcement learning (FRL), and policy distillation (PD).

Planning and reinforcement learning are two key approaches to sequential decision making. Multi-step approximate real-time dynamic programming, a recently successful algorithm class of which AlphaZero [Silver et al., 2018] is an example, combines both by nesting planning within a learning loop. However, the combination of planning and learning introduces a new question: how should we balance time spend on planning, learning and acting? The importance of this trade-off has not been explicitly studied before. We show that it is actually of key importance, with computational results indicating that we should neither plan too long nor too short. Conceptually, we identify a new spectrum of planning-learning algorithms which ranges from exhaustive search (long planning) to model-free RL (no planning), with optimal performance achieved midway.

Whenever countries are threatened by a pandemic, as is the case with the COVID-19 virus, governments should take the right actions to safeguard public health as well as to mitigate the negative effects on the economy. In this regard, there are two completely different approaches governments can take: a restrictive one, in which drastic measures such as self-isolation can seriously damage the economy, and a more liberal one, where more relaxed restrictions may put at risk a high percentage of the population. The optimal approach could be somewhere in between, and, in order to make the right decisions, it is necessary to accurately estimate the future effects of taking one or other measures. In this paper, we use the SEIR epidemiological model (Susceptible - Exposed - Infected - Recovered) for infectious diseases to represent the evolution of the virus COVID-19 over time in the population. To optimize the best sequences of actions governments can take, we propose a methodology with two approaches, one based on Deep Q-Learning and another one based on Genetic Algorithms. The sequences of actions (confinement, self-isolation, two-meter distance or not taking restrictions) are evaluated according to a reward system focused on meeting two objectives: firstly, getting few people infected so that hospitals are not overwhelmed with critical patients, and secondly, avoiding taking drastic measures for too long which can potentially cause serious damage to the economy. The conducted experiments prove that our methodology is a valid tool to discover actions governments can take to reduce the negative effects of a pandemic in both senses. We also prove that the approach based on Deep Q-Learning overcomes the one based on Genetic Algorithms for optimizing the sequences of actions.

Adversarial attacks against conventional Deep Learning (DL) systems and algorithms have been widely studied, and various defenses were proposed. However, the possibility and feasibility of such attacks against Deep Reinforcement Learning (DRL) are less explored. As DRL has achieved great success in various complex tasks, designing effective adversarial attacks is an indispensable prerequisite towards building robust DRL algorithms. In this paper, we introduce two novel adversarial attack techniques to \emph{stealthily} and \emph{efficiently} attack the DRL agents. These two techniques enable an adversary to inject adversarial samples in a minimal set of critical moments while causing the most severe damage to the agent. The first technique is the \emph{critical point attack}: the adversary builds a model to predict the future environmental states and agent's actions, assesses the damage of each possible attack strategy, and selects the optimal one. The second technique is the \emph{antagonist attack}: the adversary automatically learns a domain-agnostic model to discover the critical moments of attacking the agent in an episode. Experimental results demonstrate the effectiveness of our techniques. Specifically, to successfully attack the DRL agent, our critical point technique only requires 1 (TORCS) or 2 (Atari Pong and Breakout) steps, and the antagonist technique needs fewer than 5 steps (4 Mujoco tasks), which are significant improvements over state-of-the-art methods.

With the increasing popularity of electric vehicles, distributed energy generation and storage facilities in smart grid systems, an efficient Demand-Side Management (DSM) is urgent for energy savings and peak loads reduction. Traditional DSM works focusing on optimizing the energy activities for a single household can not scale up to large-scale home energy management problems. Multi-agent Deep Reinforcement Learning (MA-DRL) shows a potential way to solve the problem of scalability, where modern homes interact together to reduce energy consumers consumption while striking a balance between energy cost and peak loads reduction. However, it is difficult to solve such an environment with the non-stationarity, and existing MA-DRL approaches cannot effectively give incentives for expected group behavior. In this paper, we propose a collective MA-DRL algorithm with continuous action space to provide fine-grained control on a large scale microgrid. To mitigate the non-stationarity of the microgrid environment, a novel predictive model is proposed to measure the collective market behavior. Besides, a collective behavior entropy is introduced to reduce the high peak loads incurred by the collective behaviors of all householders in the smart grid. Empirical results show that our approach significantly outperforms the state-of-the-art methods regarding power cost reduction and daily peak loads optimization.

We present a self-learning approach that combines deep reinforcement learning and Monte Carlo tree search to solve the traveling salesman problem. The proposed approach has two advantages. First, it adopts deep reinforcement learning to compute the value functions for decision, which removes the need of hand-crafted features and labelled data. Second, it uses Monte Carlo tree search to select the best policy by comparing different value functions, which increases its generalization ability. Experimental results show that the proposed method performs favorably against other methods in small-to-medium problem settings. And it shows comparable performance as state-of-the-art in large problem setting.

The promise of reinforcement learning is to solve complex sequential decision problems by specifying a high-level reward function only. However, RL algorithms struggle when, as is often the case, simple and intuitive rewards provide sparse and deceptive feedback. Avoiding these pitfalls requires thoroughly exploring the environment, but despite substantial investments by the community, creating algorithms that can do so remains one of the central challenges of the field. We hypothesize that the main impediment to effective exploration originates from algorithms forgetting how to reach previously visited states ("detachment") and from failing to first return to a state before exploring from it ("derailment"). We introduce Go-Explore, a family of algorithms that addresses these two challenges directly through the simple principles of explicitly remembering promising states and first returning to such states before exploring. Go-Explore solves all heretofore unsolved Atari games (those for which algorithms could not previously outperform humans when evaluated following current community standards) and surpasses the state of the art on all hard-exploration games, with orders of magnitude improvements on the grand challenges Montezuma's Revenge and Pitfall. We also demonstrate the practical potential of Go-Explore on a challenging and extremely sparse-reward robotics task. Additionally, we show that adding a goal-conditioned policy can further improve Go-Explore's exploration efficiency and enable it to handle stochasticity throughout training. The striking contrast between the substantial performance gains from Go-Explore and the simplicity of its mechanisms suggests that remembering promising states, returning to them, and exploring from them is a powerful and general approach to exploration, an insight that may prove critical to the creation of truly intelligent learning agents.

When chasing a bouncing ball, a human will head where they anticipate the ball is going. If things change -- for example a cat swats the ball and it bounces off in a new direction -- the human will correct to an appropriate new route in real time. Robots can have a hard time making such changes, as they tend to simply observe states, then calculate and execute actions, rather than thinking while moving. Google Brain, UC Berkeley, and X Lab have proposed a concurrent Deep Reinforcement Learning (DRL) algorithm that enables robots to take a broader and more long-term view of tasks and behaviours, and decide on their next action before the current one is completed. The paper has been accepted by ICLR 2020.

Explainable Artificial Intelligence (XAI), i.e., the development of more transparent and interpretable AI models, has gained increased traction over the last few years. This is due to the fact that, in conjunction with their growth into powerful and ubiquitous tools, AI models exhibit one detrimential characteristic: a performance-transparency trade-off. This describes the fact that the more complex a model's inner workings, the less clear it is how its predictions or decisions were achieved. But, especially considering Machine Learning (ML) methods like Reinforcement Learning (RL) where the system learns autonomously, the necessity to understand the underlying reasoning for their decisions becomes apparent. Since, to the best of our knowledge, there exists no single work offering an overview of Explainable Reinforcement Learning (XRL) methods, this survey attempts to address this gap. We give a short summary of the problem, a definition of important terms, and offer a classification and assessment of current XRL methods. We found that a) the majority of XRL methods function by mimicking and simplifying a complex model instead of designing an inherently simple one, and b) XRL (and XAI) methods often neglect to consider the human side of the equation, not taking into account research from related fields like psychology or philosophy. Thus, an interdisciplinary effort is needed to adapt the generated explanations to a (non-expert) human user in order to effectively progress in the field of XRL and XAI in general.