In this report, we summarize the takeaways from the first NeurIPS 2021 NetHack Challenge. Participants were tasked with developing a program or agent that can win (i.e., 'ascend' in) the popular dungeon-crawler game of NetHack by interacting with the NetHack Learning Environment (NLE), a scalable, procedurally generated, and challenging Gym environment for reinforcement learning (RL). The challenge showcased community-driven progress in AI with many diverse approaches significantly beating the previously best results on NetHack. Furthermore, it served as a direct comparison between neural (e.g., deep RL) and symbolic AI, as well as hybrid systems, demonstrating that on NetHack symbolic bots currently outperform deep RL by a large margin. Lastly, no agent got close to winning the game, illustrating NetHack's suitability as a long-term benchmark for AI research.

Q-learning is one of the most simple yet popular algorithms in the reinforcement learning (RL) community [Sutton and Barto, 2018]. However, Q-learning suffers the divergence issue when (linear) function approximation is applied [Baird, 1995, Tsitsiklis and Van Roy, 1997]. To address this instability issue, a technique called target network is proposed in the famous DQN algorithm [Mnih et al., 2015]. In particular, DQN implements a duplication of the main Q-network (i.e., the so-called target network), which is further used to generate the bootstrap signal for updates. One important feature is that the target network is fixed over intervals. Unlike Q-learning, the learning targets do not change during an interval for DQN. In [Mnih et al., 2015, Table 3], it is reported that the target network contributes a lot to the superior performance of DQN.

Significant advances have recently been achieved in Multi-Agent Reinforcement Learning (MARL) which tackles sequential decision-making problems involving multiple participants. However, MARL requires a tremendous number of samples for effective training. On the other hand, model-based methods have been shown to achieve provable advantages of sample efficiency. However, the attempts of model-based methods to MARL have just started very recently. This paper presents a review of the existing research on model-based MARL, including theoretical analyses, algorithms, and applications, and analyzes the advantages and potential of model-based MARL. Specifically, we provide a detailed taxonomy of the algorithms and point out the pros and cons for each algorithm according to the challenges inherent to multi-agent scenarios. We also outline promising directions for future development of this field.

This paper presents a new reward function that can be used for deep reinforcement learning in unmanned aerial vehicle (UAV) control and navigation problems. The reward function is based on the construction and estimation of the time of simplified trajectories to the target, which are third-order Bezier curves. This reward function can be applied unchanged to solve problems in both two-dimensional and three-dimensional virtual environments. The effectiveness of the reward function was tested in a newly developed virtual environment, namely, a simplified two-dimensional environment describing the dynamics of UAV control and flight, taking into account the forces of thrust, inertia, gravity, and aerodynamic drag. In this formulation, three tasks of UAV control and navigation were successfully solved: UAV flight to a given point in space, avoidance of interception by another UAV, and organization of interception of one UAV by another. The three most relevant modern deep reinforcement learning algorithms, Soft actor-critic, Deep Deterministic Policy Gradient, and Twin Delayed Deep Deterministic Policy Gradient were used. All three algorithms performed well, indicating the effectiveness of the selected reward function.

Deep reinforcement learning (DRL) is one of the promising approaches for introducing robots into complicated environments. The recent remarkable progress of DRL stands on regularization of policy, which allows the policy to improve stably and efficiently. A popular method, so-called proximal policy optimization (PPO), and its variants constrain density ratio of the latest and baseline policies when the density ratio exceeds a given threshold. This threshold can be designed relatively intuitively, and in fact its recommended value range has been suggested. However, the density ratio is asymmetric for its center, and the possible error scale from its center, which should be close to the threshold, would depend on how the baseline policy is given. In order to maximize the values of regularization of policy, this paper proposes a new PPO derived using relative Pearson (RPE) divergence, therefore so-called PPO-RPE, to design the threshold adaptively. In PPO-RPE, the relative density ratio, which can be formed with symmetry, replaces the raw density ratio. Thanks to this symmetry, its error scale from center can easily be estimated, hence, the threshold can be adapted for the estimated error scale. From three simple benchmark simulations, the importance of algorithm-dependent threshold design is revealed. By simulating additional four locomotion tasks, it is verified that the proposed method statistically contributes to task accomplishment by appropriately restricting the policy updates.

Reinforcement learning (RL) approaches can illuminate emergent behaviors that facilitate coordination across teams of agents as part of a multi-agent system (MAS), which can provide windows of opportunity in various military tasks. Technologically advancing adversaries pose substantial risks to a friendly nation's interests and resources. Superior resources alone are not enough to defeat adversaries in modern complex environments because adversaries create standoff in multiple domains against predictable military doctrine-based maneuvers. Therefore, as part of a defense strategy, friendly forces must use strategic maneuvers and disruption to gain superiority in complex multi-faceted domains such as multi-domain operations (MDO). One promising avenue for implementing strategic maneuver and disruption to gain superiority over adversaries is through coordination of MAS in future military operations. In this paper, we present overviews of prominent works in the RL domain with their strengths and weaknesses for overcoming the challenges associated with performing autonomous strategic maneuver and disruption in military contexts.

Asynchronous and parallel implementation of standard reinforcement learning (RL) algorithms is a key enabler of the tremendous success of modern RL. Among many asynchronous RL algorithms, arguably the most popular and effective one is the asynchronous advantage actor-critic (A3C) algorithm. Although A3C is becoming the workhorse of RL, its theoretical properties are still not well-understood, including its non-asymptotic analysis and the performance gain of parallelism (a.k.a. linear speedup). This paper revisits the A3C algorithm and establishes its non-asymptotic convergence guarantees. Under both i.i.d. and Markovian sampling, we establish the local convergence guarantee for A3C in the general policy approximation case and the global convergence guarantee in softmax policy parameterization. Under i.i.d. sampling, A3C obtains sample complexity of $\mathcal{O}(\epsilon^{-2.5}/N)$ per worker to achieve $\epsilon$ accuracy, where $N$ is the number of workers. Compared to the best-known sample complexity of $\mathcal{O}(\epsilon^{-2.5})$ for two-timescale AC, A3C achieves \emph{linear speedup}, which justifies the advantage of parallelism and asynchrony in AC algorithms theoretically for the first time. Numerical tests on synthetic environment, OpenAI Gym environments and Atari games have been provided to verify our theoretical analysis.

Simulation and synthesis are core parts of the future of AI and machine learning. Consider: programmers, data scientists, and machine learning engineers can create the brain of a self-driving car without the car. Rather than use information from the real world, you can create artificial data using simulations to train traditional machine learning models. With this practical book, you'll explore the possibilities of simulation- and synthesis-based machine learning and AI, with a focus on deep reinforcement learning and imitation learning techniques. AI and ML are increasingly data driven, and simulations are a powerful, engaging way to unlock their full potential.

Prosumer operators are dealing with extensive challenges to participate in short-term electricity markets while taking uncertainties into account. Challenges such as variation in demand, solar energy, wind power, and electricity prices as well as faster response time in intraday electricity markets. Machine learning approaches could resolve these challenges due to their ability to continuous learning of complex relations and providing a real-time response. Such approaches are applicable with presence of the high performance computing and big data. To tackle these challenges, a Markov decision process is proposed and solved with a reinforcement learning algorithm with proper observations and actions employing tabular Q-learning. Trained agent converges to a policy which is similar to the global optimal solution. It increases the prosumer's profit by 13.39% compared to the well-known stochastic optimization approach.

We study two state-of-the-art solutions to the multi-agent pickup and delivery (MAPD) problem based on different principles -- multi-agent path-finding (MAPF) and multi-agent reinforcement learning (MARL). Specifically, a recent MAPF algorithm called conflict-based search (CBS) and a current MARL algorithm called shared experience actor-critic (SEAC) are studied. While the performance of these algorithms is measured using quite different metrics in their separate lines of work, we aim to benchmark these two methods comprehensively in a simulated warehouse automation environment.