Counterfactual regret minimization (CFR) is a family of algorithms for effectively solving imperfect-information games. To enhance CFR's applicability in large games, researchers use neural networks to approximate its behavior. However, existing methods are mainly based on vanilla CFR and struggle to effectively integrate more advanced CFR variants. In this work, we propose an efficient model-free neural CFR algorithm, overcoming the limitations of existing methods in approximating advanced CFR variants. At each iteration, it collects variance-reduced sampled advantages based on a value network, fits cumulative advantages by bootstrapping, and applies discounting and clipping operations to simulate the update mechanisms of advanced CFR variants. Experimental results show that, compared with model-free neural algorithms, it exhibits faster convergence in typical imperfect-information games and demonstrates stronger adversarial performance in a large poker game.

Opponent Modeling (OM) aims to enhance decision-making by modeling other agents in multi-agent environments. Existing works typically learn opponent models against a pre-designated fixed set of opponents during training. However, this will cause poor generalization when facing unknown opponents during testing, as previously unseen opponents can exhibit out-of-distribution (OOD) behaviors that the learned opponent models cannot handle. To tackle this problem, we introduce a novel Open-Ended Opponent Modeling (OEOM) framework, which continuously generates opponents with diverse strengths and styles to reduce the possibility of OOD situations occurring during testing. Founded on population-based training and information-theoretic trajectory space diversity regularization, OEOM generates a dynamic set of opponents. This set is then fed to any OM approaches to train a potentially generalizable opponent model. Upon this, we further propose a simple yet effective OM approach that naturally fits within the OEOM framework. This approach is based on in-context reinforcement learning and learns a Transformer that dynamically recognizes and responds to opponents based on their trajectories. Extensive experiments in cooperative, competitive, and mixed environments demonstrate that OEOM is an approach-agnostic framework that improves generalizability compared to training against a fixed set of opponents, regardless of OM approaches or testing opponent settings. The results also indicate that our proposed approach generally outperforms existing OM baselines.

Recovering a spectrum of diverse policies from a set of expert trajectories is an important research topic in imitation learning. After determining a latent style for a trajectory, previous diverse polices recovering methods usually employ a vanilla behavioral cloning learning objective conditioned on the latent style, treating each state-action pair in the trajectory with equal importance. Based on an observation that in many scenarios, behavioral styles are often highly relevant with only a subset of state-action pairs, this paper presents a new principled method in diverse polices recovering. In particular, after inferring or assigning a latent style for a trajectory, we enhance the vanilla behavioral cloning by incorporating a weighting mechanism based on pointwise mutual information. This additional weighting reflects the significance of each state-action pair's contribution to learning the style, thus allowing our method to focus on state-action pairs most representative of that style. We provide theoretical justifications for our new objective, and extensive empirical evaluations confirm the effectiveness of our method in recovering diverse polices from expert data.

Offline multi-task reinforcement learning aims to learn a unified policy capable of solving multiple tasks using only pre-collected task-mixed datasets, without requiring any online interaction with the environment. However, it faces significant challenges in effectively sharing knowledge across tasks. Inspired by the efficient knowledge abstraction observed in human learning, we propose Goal-Oriented Skill Abstraction (GO-Skill), a novel approach designed to extract and utilize reusable skills to enhance knowledge transfer and task performance. Our approach uncovers reusable skills through a goal-oriented skill extraction process and leverages vector quantization to construct a discrete skill library. To mitigate class imbalances between broadly applicable and task-specific skills, we introduce a skill enhancement phase to refine the extracted skills. Furthermore, we integrate these skills using hierarchical policy learning, enabling the construction of a high-level policy that dynamically orchestrates discrete skills to accomplish specific tasks. Extensive experiments on diverse robotic manipulation tasks within the MetaWorld benchmark demonstrate the effectiveness and versatility of GO-Skill.

Offline Opponent Modeling (OOM) aims to learn an adaptive autonomous agent policy that dynamically adapts to opponents using an offline dataset from multi-agent games. Previous work assumes that the dataset is optimal. However, this assumption is difficult to satisfy in the real world. When the dataset is suboptimal, existing approaches struggle to work. To tackle this issue, we propose a simple and general algorithmic improvement framework, Truncated Q-driven Instant Policy Refinement (TIPR), to handle the suboptimality of OOM algorithms induced by datasets. The TIPR framework is plug-and-play in nature. Compared to original OOM algorithms, it requires only two extra steps: (1) Learn a horizon-truncated in-context action-value function, namely Truncated Q, using the offline dataset. The Truncated Q estimates the expected return within a fixed, truncated horizon and is conditioned on opponent information. (2) Use the learned Truncated Q to instantly decide whether to perform policy refinement and to generate policy after refinement during testing. Theoretically, we analyze the rationale of Truncated Q from the perspective of No Maximization Bias probability. Empirically, we conduct extensive comparison and ablation experiments in four representative competitive environments. TIPR effectively improves various OOM algorithms pretrained with suboptimal datasets.

Developing AI agents capable of interacting with open-world environments to solve diverse tasks is a compelling challenge. However, evaluating such open-ended agents remains difficult, with current benchmarks facing scalability limitations. To address this, we introduce \textit{Minecraft Universe} (MCU), a comprehensive evaluation framework set within the open-world video game Minecraft. MCU incorporates three key components: (1) an expanding collection of 3,452 composable atomic tasks that encompasses 11 major categories and 41 subcategories of challenges; (2) a task composition mechanism capable of generating infinite diverse tasks with varying difficulty; and (3) a general evaluation framework that achieves 91.5\% alignment with human ratings for open-ended task assessment. Empirical results reveal that even state-of-the-art foundation agents struggle with the increasing diversity and complexity of tasks. These findings highlight the necessity of MCU as a robust benchmark to drive progress in AI agent development within open-ended environments. Our evaluation code and scripts are available at https://github.com/CraftJarvis/MCU.

Despite significant advances in learning imperfect information extensive-form games (IIEFGs), most existing theoretical guarantees are limited to IIEFGs in the tabular case. To permit efficient learning of large-scale IIEFGs, we take the first step in studying two-player zero-sum IIEFGs with linear function approximation. In particular, we consider linear IIEFGs in the formulation of partially observable Markov games (POMGs) with linearly parameterized rewards. To address the challenge that the underlying function approximation structure is difficult to directly apply due to the imperfect information of states, we construct the composite "feature vectors" for information set-action pairs. Based on this, we further propose a "least-squares loss estimator", which we call the *fictitious* least-squares loss estimator. Through integrating this estimator with the follow-the-regularized-leader (FTRL) framework, we propose the *fictitious* least-squares follow-the-regularized-leader ($\text{F}^2\text{TRL}$) algorithm, which achieves a provable $\widetilde{\mathcal{O}}(\lambda\sqrt{d H^2 T})$ regret guarantee in the large $T$ regime, where $d$ is the ambient dimension of the feature mapping, $H$ is the horizon length, $\lambda$ is a "balance coefficient" and $T$ is the number of episodes. At the core of the analysis of $\text{F}^2\text{TRL}$ is the leverage of our proposed new "balanced transition" over information set-action space. Additionally, we complement our results with an $\Omega(\sqrt{d\min(d,H)T})$ regret lower bound for this problem and conduct empirical evaluations across various environments, which corroborate the effectiveness of our algorithm.

Generating agents that can achieve zero-shot coordination (ZSC) with unseen partners is a new challenge in cooperative multiagent reinforcement learning (MARL). Recently, some studies have made progress in ZSC by exposing the agents to diverse partners during the training process. They usually involve self-play when training the partners, implicitly assuming that the tasks are homogeneous. However, many real-world tasks are heterogeneous, and hence previous methods may be inefficient. In this article, we study the heterogeneous ZSC problem for the first time and propose a general method based on coevolution, which coevolves two populations of agents and partners through three subprocesses: 1) pairing; 2) updating; and 3) selection. Experimental results on various heterogeneous tasks highlight the necessity of considering the heterogeneous setting and demonstrate that our proposed method is a promising solution for heterogeneous ZSC tasks. To the best of our knowledge, we are the first to underscore the significance of the heterogeneous ZSC tasks and to introduce an effective framework for addressing it.

Multi-task reinforcement learning endeavors to accomplish a set of different tasks with a single policy. To enhance data efficiency by sharing parameters across multiple tasks, a common practice segments the network into distinct modules and trains a routing network to recombine these modules into task-specific policies. However, existing routing approaches employ a fixed number of modules for all tasks, neglecting that tasks with varying difficulties commonly require varying amounts of knowledge. This work presents a Dynamic Depth Routing (D2R) framework, which learns strategic skipping of certain intermediate modules, thereby flexibly choosing different numbers of modules for each task. Under this framework, we further introduce a ResRouting method to address the issue of disparate routing paths between behavior and target policies during off-policy training. In addition, we design an automatic route-balancing mechanism to encourage continued routing exploration for unmastered tasks without disturbing the routing of mastered ones. We conduct extensive experiments on various robotics manipulation tasks in the Meta-World benchmark, where D2R achieves state-of-the-art performance with significantly improved learning efficiency.

Building large-scale generalist pre-trained models for many tasks is becoming an emerging and potential direction in reinforcement learning (RL).Research such as Gato and Multi-Game Decision Transformer have displayed outstanding performance and generalization capabilities on many games and domains.However, there exists a research blank about developing highly capable and generalist models in multi-agent RL (MARL), which can substantially accelerate progress toward general AI.To fill this gap, we propose Multi-Agent multi-Game ENtity TrAnsformer (MA-GENTA) from the entity perspective as orthogonal research to previous time-sequential modeling.Specifically, to deal with different state/observation spaces in different games, we analogize games as languages by aligning one single game to one single language, thus training different "tokenizers" and a shared transformer for various games.The feature inputs are split according to different entities and tokenized in the same continuous space.Then, two types of transformer-based models are proposed as permutationinvariant architectures to deal with various numbers of entities and capture the attention of different entities.MAGENTA is trained on

Existing game AI research mainly focuses on enhancing agents' abilities to win games, but this does not inherently make humans have a better experience when collaborating with these agents. For example, agents may dominate the collaboration and exhibit unintended or detrimental behaviors, leading to poor experiences for their human partners. In other words, most game AI agents are modeled in a "self-centered" manner. In this paper, we propose a "human-centered" modeling scheme for collaborative agents that aims to enhance the experience of humans. Specifically, we model the experience of humans as the goals they expect to achieve during the task. We expect that agents should learn to enhance the extent to which humans achieve these goals while maintaining agents' original abilities (e.g., winning games). To achieve this, we propose the Reinforcement Learning from Human Gain (RLHG) approach. The RLHG approach introduces a "baseline", which corresponds to the extent to which humans primitively achieve their goals, and encourages agents to learn behaviors that can effectively enhance humans in achieving their goals better. We evaluate the RLHG agent in the popular Multi-player Online Battle Arena (MOBA) game, Honor of Kings, by conducting real-world human-agent tests. Both objective performance and subjective preference results show that the RLHG agent provides participants better gaming experience.

Opponent modeling aims at learning the opponent's behaviors, goals, or beliefs to reduce the uncertainty of the competitive environment and assist decision-making. Existing work has mostly focused on learning opponent models online, which is impractical and inefficient in practical scenarios. To this end, we formalize an Offline Opponent Modeling (OOM) problem with the objective of utilizing pre-collected offline datasets to learn opponent models that characterize the opponent from the viewpoint of the controlled agent, which aids in adapting to the unknown fixed policies of the opponent. Drawing on the promises of the Transformers for decision-making, we introduce a general approach, Transformer Against Opponent (TAO), for OOM. Essentially, TAO tackles the problem by harnessing the full potential of the supervised pre-trained Transformers' in-context learning capabilities. The foundation of TAO lies in three stages: an innovative offline policy embedding learning stage, an offline opponent-aware response policy training stage, and a deployment stage for opponent adaptation with in-context learning. Theoretical analysis establishes TAO's equivalence to Bayesian posterior sampling in opponent modeling and guarantees TAO's convergence in opponent policy recognition. Extensive experiments and ablation studies on competitive environments with sparse and dense rewards demonstrate the impressive performance of TAO. Our approach manifests remarkable prowess for fast adaptation, especially in the face of unseen opponent policies, confirming its in-context learning potency.

*Multi-agent reinforcement learning* (MARL) has been shown effective for cooperative games in recent years. However, existing state-of-the-art methods face challenges related to sample complexity, training instability, and the risk of converging to a suboptimal Nash Equilibrium. In this paper, we propose a unified framework for learning \emph{stochastic} policies to resolve these issues. We embed cooperative MARL problems into probabilistic graphical models, from which we derive the maximum entropy (MaxEnt) objective for MARL. Based on the MaxEnt framework, we propose *Heterogeneous-Agent Soft Actor-Critic* (HASAC) algorithm. Theoretically, we prove the monotonic improvement and convergence to *quantal response equilibrium* (QRE) properties of HASAC. Furthermore, we generalize a unified template for MaxEnt algorithmic design named *Maximum Entropy Heterogeneous-Agent Mirror Learning* (MEHAML), which provides any induced method with the same guarantees as HASAC. We evaluate HASAC on six benchmarks: Bi-DexHands, Multi-Agent MuJoCo, StarCraft Multi-Agent Challenge, Google Research Football, Multi-Agent Particle Environment, and Light Aircraft Game. Results show that HASAC consistently outperforms strong baselines, exhibiting better sample efficiency, robustness, and sufficient exploration.

Counterfactual regret minimization (CFR) is a family of iterative algorithms showing promising results in solving imperfect-information games. Recent novel CFR variants (e.g., CFR+, DCFR) have significantly improved the convergence rate of the vanilla CFR. The key to these CFR variants’ performance is weighting each iteration non-uniformly, i.e., discounting earlier iterations. However, these algorithms use a fixed, manually-specified scheme to weight each iteration, which enormously limits their potential. In this work, we propose Dynamic Discounted CFR (DDCFR), the first equilibrium-finding framework that discounts prior iterations using a dynamic, automatically-learned scheme. We formalize CFR’s iteration process as a carefully designed Markov decision process and transform the discounting scheme learning problem into a policy optimization problem within it. The learned discounting scheme dynamically weights each iteration on the fly using information available at runtime. Experimental results across multiple games demonstrate that DDCFR’s dynamic discounting scheme has a strong generalization ability and leads to faster convergence with improved performance. The code is available at https://github.com/rpSebastian/DDCFR.

Multi-task reinforcement learning endeavors to efficiently leverage shared information across various tasks, facilitating the simultaneous learning of multiple tasks. Existing approaches primarily focus on parameter sharing with carefully designed network structures or tailored optimization procedures. However, they overlook a direct and complementary way to exploit cross-task similarities: the control policies of tasks already proficient in some skills can provide explicit guidance for unmastered tasks to accelerate skills acquisition. To this end, we present a novel framework called Cross-Task Policy Guidance (CTPG), which trains a guide policy for each task to select the behavior policy interacting with the environment from all tasks' control policies, generating better training trajectories. In addition, we propose two gating mechanisms to improve the learning efficiency of CTPG: one gate filters out control policies that are not beneficial for guidance, while the other gate blocks tasks that do not necessitate guidance. CTPG is a general framework adaptable to existing parameter sharing approaches. Empirical evaluations demonstrate that incorporating CTPG with these approaches significantly enhances performance in manipulation and locomotion benchmarks.

Opponent modeling is a longstanding research topic aimed at enhancing decision-making by modeling information about opponents in multi-agent environments. However, existing approaches often face challenges such as having difficulty generalizing to unknown opponent policies and conducting unstable performance. To tackle these challenges, we propose a novel approach based on in-context learning and decision-time search named Opponent Modeling with In-context Search (OMIS). OMIS leverages in-context learning-based pretraining to train a Transformer model for decision-making. It consists of three in-context components: an actor learning best responses to opponent policies, an opponent imitator mimicking opponent actions, and a critic estimating state values. When testing in an environment that features unknown non-stationary opponent agents, OMIS uses pretrained in-context components for decision-time search to refine the actor's policy. Theoretically, we prove that under reasonable assumptions, OMIS without search converges in opponent policy recognition and has good generalization properties; with search, OMIS provides improvement guarantees, exhibiting performance stability. Empirically, in competitive, cooperative, and mixed environments, OMIS demonstrates more effective and stable adaptation to opponents than other approaches. See our project website at https://sites.google.com/view/nips2024-omis.

A key challenge of continual reinforcement learning (CRL) in dynamic environments is to promptly adapt the reinforcement learning (RL) agent's behavior as the environment changes over its lifetime while minimizing the catastrophic forgetting of the learned information. To address this challenge, in this article, we propose DaCoRL, that is, dynamics-adaptive continual RL. DaCoRL learns a context-conditioned policy using progressive contextualization, which incrementally clusters a stream of stationary tasks in the dynamic environment into a series of contexts and opts for an expandable multihead neural network to approximate the policy. Specifically, we define a set of tasks with similar dynamics as an environmental context and formalize context inference as a procedure of online Bayesian infinite Gaussian mixture clustering on environment features, resorting to online Bayesian inference to infer the posterior distribution over contexts. Under the assumption of a Chinese restaurant process (CRP) prior, this technique can accurately classify the current task as a previously seen context or instantiate a new context as needed without relying on any external indicator to signal environmental changes in advance. Furthermore, we employ an expandable multihead neural network whose output layer is synchronously expanded with the newly instantiated context and a knowledge distillation regularization term for retaining the performance on learned tasks. As a general framework that can be coupled with various deep RL algorithms, DaCoRL features consistent superiority over existing methods in terms of stability, overall performance, and generalization ability, as verified by extensive experiments on several robot navigation and MuJoCo locomotion tasks.

Whole-slide images (WSI) in computational pathology have high resolution with gigapixel size, but are generally with sparse regions of interest, which leads to weak diagnostic relevance and data inefficiency for each area in the slide. Most of the existing methods rely on a multiple instance learning framework that requires densely sampling local patches at high magnification. The limitation is evident in the application stage as the heavy computation for extracting patch-level features is inevitable. In this paper, we develop RLogist, a benchmarking deep reinforcement learning (DRL) method for fast observation strategy on WSIs. Imitating the diagnostic logic of human pathologists, our RL agent learns how to find regions of observation value and obtain representative features across multiple resolution levels, without having to analyze each part of the WSI at the high magnification. We benchmark our method on two whole-slide level classification tasks, including detection of metastases in WSIs of lymph node sections, and subtyping of lung cancer. Experimental results demonstrate that RLogist achieves competitive classification performance compared to typical multiple instance learning algorithms, while having a significantly short observation path. In addition, the observation path given by RLogist provides good decision-making interpretability, and its ability of reading path navigation can potentially be used by pathologists for educational/assistive purposes. Our code is available at: https://github.com/tencent-ailab/RLogist.

In recent years, online search has been playing an increasingly important role in imperfect information games (IIGs). Previous online search is known as common-knowledge subgame solving, which has to consider all the states in a common-knowledge closure. This is only computationally tolerable for medium size games, such as poker. To handle larger games, order-1 Knowledge-Limited Subgame Solving (1-KLSS) only considers the states in a knowledge-limited closure, which results in a much smaller subgame. However, 1-KLSS is unsafe. In this paper, we first extend 1-KLSS to Safe-1-KLSS and prove its safeness. To make Safe-1-KLSS applicable to even larger games, we propose Opponent-Limited Subgame Solving (OLSS) to limit how the opponent reaches a subgame and how it acts in the subgame. Limiting the opponent's strategy dramatically reduces the subgame size and improves the efficiency of subgame solving while still preserving some safety in the limit. Experiments in medium size poker show that Safe-1-KLSS and OLSS are orders of magnitude faster than previous common-knowledge subgame solving. Also, OLSS significantly improves the online performance in a two-player Mahjong game, whose game size prohibits the use of previous common-knowledge subgame-solving methods.

Recent research in offline reinforcement learning (RL) has demonstrated that return-conditioned supervised learning is a powerful paradigm for decision-making problems. While promising, return conditioning is limited to training data labeled with rewards and therefore faces challenges in learning from unsupervised data. In this work, we aim to utilize generalized future conditioning to enable efficient unsupervised pretraining from reward-free and sub-optimal offline data. We propose Pretrained Decision Transformer (PDT), a conceptually simple approach for unsupervised RL pretraining. PDT leverages future trajectory information as a privileged context to predict actions during training. The ability to make decisions based on both present and future factors enhances PDT's capability for generalization. Besides, this feature can be easily incorporated into a return-conditioned framework for online finetuning, by assigning return values to possible futures and sampling future embeddings based on their respective values. Empirically, PDT outperforms or performs on par with its supervised pretraining counterpart, especially when dealing with sub-optimal data. Further analysis reveals that PDT can extract diverse behaviors from offline data and controllably sample high-return behaviors by online finetuning. Code is available at here.

Quality-Diversity (QD) algorithms, a subset of evolutionary algorithms, maintain an archive (i.e., a set of solutions) and simulate the natural evolution process through iterative selection and reproduction, with the goal of generating a set of high-quality and diverse solutions. Though having found many successful applications in reinforcement learning, QD algorithms often select the parent solutions uniformly at random, which lacks selection pressure and may limit the performance. Recent studies have treated each type of behavior of a solution as an objective, and selected the parent solutions based on Multi-objective Optimization (MO), which is a natural idea, but has not lead to satisfactory performance as expected. This paper gives the reason for the first time, and then proposes a new MO-based selection method by non-surrounded-dominated sorting (NSS), which considers all possible directions of the behaviors, and thus can generate diverse solutions over the whole behavior space. By combining NSS with the most widespread QD algorithm, MAP-Elites, we perform experiments on synthetic functions and several complex tasks (i.e., QDGym, robotic arm, and Mario environment generation), showing that NSS achieves better performance than not only other MO-based selection methods but also state-of-the-art selection methods in QD.

It is extremely difficult to train a superhuman Artificial Intelligence (AI) for games of similar size to StarCraft II. AlphaStar is the first AI that beat human professionals in the full game of StarCraft II, using a league training framework that is inspired by a game-theoretic approach. In this paper, we improve AlphaStar's league training in two significant aspects. We train goal-conditioned exploiters, whose abilities of spotting weaknesses in the main agent and the entire league are greatly improved compared to the unconditioned exploiters in AlphaStar. In addition, we endow the agents in the league with the new ability of opponent modeling, which makes the agent more responsive to the opponent's real-time strategy. Based on these improvements, we train a better and superhuman AI with orders of magnitude less resources than AlphaStar (see Table 1 for a full comparison). Considering the iconic role of StarCraft II in game AI research, we believe our method and results on StarCraft II provide valuable design principles on how one would utilize the general league training framework for obtaining a least-exploitable strategy in various, large-scale, real-world games.

The advancement of Offline Reinforcement Learning (RL) and Offline Multi-Agent Reinforcement Learning (MARL) critically depends on the availability of high-quality, pre-collected offline datasets that represent real-world complexities and practical applications. However, existing datasets often fall short in their simplicity and lack of realism. To address this gap, we propose Hokoff, a comprehensive set of pre-collected datasets that covers both offline RL and offline MARL, accompanied by a robust framework, to facilitate further research. This data is derived from Honor of Kings, a recognized Multiplayer Online Battle Arena (MOBA) game known for its intricate nature, closely resembling real-life situations. Utilizing this framework, we benchmark a variety of offline RL and offline MARL algorithms. We also introduce a novel baseline algorithm tailored for the inherent hierarchical action space of the game. We reveal the incompetency of current offline RL approaches in handling task complexity, generalization and multi-task learning.

Policy-Space Response Oracles (PSRO) is an influential algorithm framework for approximating a Nash Equilibrium (NE) in multi-agent non-transitive games. Many previous studies have been trying to promote policy diversity in PSRO. A major weakness with existing diversity metrics is that a more diverse (according to their diversity metrics) population does not necessarily mean (as we proved in the paper) a better approximation to a NE. To alleviate this problem, we propose a new diversity metric, the improvement of which guarantees a better approximation to a NE. Meanwhile, we develop a practical and well-justified method to optimize our diversity metric using only state-action samples. By incorporating our diversity regularization into the best response solving of PSRO, we obtain a new PSRO variant, \textit{Policy Space Diversity} PSRO (PSD-PSRO). We present the convergence property of PSD-PSRO. Empirically, extensive experiments on single-state games, Leduc, and Goofspiel demonstrate that PSD-PSRO is more effective in producing significantly less exploitable policies than state-of-the-art PSRO variants.

Grouping is ubiquitous in natural systems and is essential for promoting efficiency in team coordination. This paper proposes a novel formulation of Group-oriented Multi-Agent Reinforcement Learning (GoMARL), which learns automatic grouping without domain knowledge for efficient cooperation. In contrast to existing approaches that attempt to directly learn the complex relationship between the joint action-values and individual utilities, we empower subgroups as a bridge to model the connection between small sets of agents and encourage cooperation among them, thereby improving the learning efficiency of the whole team. In particular, we factorize the joint action-values as a combination of group-wise values, which guide agents to improve their policies in a fine-grained fashion. We present an automatic grouping mechanism to generate dynamic groups and group action-values. We further introduce a hierarchical control for policy learning that drives the agents in the same group to specialize in similar policies and possess diverse strategies for various groups. Experiments on the StarCraft II micromanagement tasks and Google Research Football scenarios verify our method's effectiveness. Extensive component studies show how grouping works and enhances performance.

Counterfactual regret minimization (CFR) is the most commonly used algorithm to approximately solving two-player zero-sum imperfect-information games (IIGs). In recent years, a series of novel CFR variants such as CFR+, Linear CFR, DCFR have been proposed and have significantly improved the convergence rate of the vanilla CFR. However, most of these new variants are hand-designed by researchers through trial and error based on different motivations, which generally requires a tremendous amount of efforts and insights. This work proposes to meta-learn novel CFR algorithms through evolution to ease the burden of manual algorithm design. We first design a search language that is rich enough to represent many existing hand-designed CFR variants. We then exploit a scalable regularized evolution algorithm with a bag of acceleration techniques to efficiently search over the combinatorial space of algorithms defined by this language. The learned novel CFR algorithm can generalize to new IIGs not seen during training and performs on par with or better than existing state-of-the-art CFR variants. The code is available at https://github.com/rpSebastian/AutoCFR.

Despite significant breakthroughs in developing gaming artificial intelligence (AI), Mahjong remains quite challenging as a popular multi-player imperfect information game. Compared with games such as Go and Texas Hold’em, Mahjong has much more invisible information, unfixed game order, and a complicated scoring system, resulting in high randomness and variance of the rewarding signals during the reinforcement learning process. This paper presents a Mahjong AI by introducing Reward Variance Reduction (RVR) into a new self-play deep reinforcement learning algorithm. RVR handles the invisibility via a relative value network which leverages the global information to guide the model to converge to the optimal strategy under an oracle with perfect information. Moreover, RVR improves the training stability using an expected reward network to adapt to the complex, dynamic, and highly stochastic reward environment. Extensive experimental results show that RVR significantly reduces the variance in Mahjong AI training and improves the model performance. After only three days of self-play training on a single server with 8 GPUs, RVR defeats 62.5% opponents on the Botzone platform.

We develop a new approach, named Greedy when Sure and Conservative when Uncertain (GSCU), to competing online against unknown and nonstationary opponents. GSCU improves in four aspects: 1) introduces a novel way of learning opponent policy embeddings offline; 2) trains offline a single best response (conditional additionally on our opponent policy embedding) instead of a finite set of separate best responses against any opponent; 3) computes online a posterior of the current opponent policy embedding, without making the discrete and ineffective decision which type the current opponent belongs to; and 4) selects online between a real-time greedy policy and a fixed conservative policy via an adversarial bandit algorithm, gaining a theoretically better regret than adhering to either. Experimental studies on popular benchmarks demonstrate GSCU’s superiority over the state-of-the-art methods. The code is available online at \url{https://github.com/YeTianJHU/GSCU}.

Learning rational behaviors in open-world games like Minecraft remains to be challenging for Reinforcement Learning (RL) research due to the compound challenge of partial observability, high-dimensional visual perception and delayed reward. To address this, we propose JueWu-MC, a sample-efficient hierarchical RL approach equipped with representation learning and imitation learning to deal with perception and exploration. Specifically, our approach includes two levels of hierarchy, where the high-level controller learns a policy to control over options and the low-level workers learn to solve each sub-task. To boost the learning of sub-tasks, we propose a combination of techniques including 1) action-aware representation learning which captures underlying relations between action and representation, 2) discriminator-based self-imitation learning for efficient exploration, and 3) ensemble behavior cloning with consistency filtering for policy robustness. Extensive experiments show that JueWu-MC significantly improves sample efficiency and outperforms a set of baselines by a large margin. Notably, we won the championship of the NeurIPS MineRL 2021 research competition and achieved the highest performance score ever.

This paper introduces Honor of Kings Arena, a reinforcement learning (RL) environment based on the Honor of Kings, one of the world’s most popular games at present. Compared to other environments studied in most previous work, ours presents new generalization challenges for competitive reinforcement learning. It is a multi-agent problem with one agent competing against its opponent; and it requires the generalization ability as it has diverse targets to control and diverse opponents to compete with. We describe the observation, action, and reward specifications for the Honor of Kings domain and provide an open-source Python-based interface for communicating with the game engine. We provide twenty target heroes with a variety of tasks in Honor of Kings Arena and present initial baseline results for RL-based methods with feasible computing resources. Finally, we showcase the generalization challenges imposed by Honor of Kings Arena and possible remedies to the challenges. All of the software, including the environment-class, are publicly available.

We present JueWu-SL, the first supervised-learning-based artificial intelligence (AI) program that achieves human-level performance in playing multiplayer online battle arena (MOBA) games. Unlike prior attempts, we integrate the macro-strategy and the micromanagement of MOBA-game-playing into neural networks in a supervised and end-to-end manner. Tested on Honor of Kings, the most popular MOBA at present, our AI performs competitively at the level of High King players in standard 5v5 games.

Offline reinforcement learning (RL) tries to learn the near-optimal policy with recorded offline experience without online exploration.Current offline RL research includes: 1) generative modeling, i.e., approximating a policy using fixed data; and 2) learning the state-action value function. While most research focuses on the state-action function part through reducing the bootstrapping error in value function approximation induced by the distribution shift of training data, the effects of error propagation in generative modeling have been neglected. In this paper, we analyze the error in generative modeling. We propose AQL (action-conditioned Q-learning), a residual generative model to reduce policy approximation error for offline RL. We show that our method can learn more accurate policy approximations in different benchmark datasets. In addition, we show that the proposed offline RL method can learn more competitive AI agents in complex control tasks under the multiplayer online battle arena (MOBA) game, Honor of Kings.

AlphaZero has achieved superhuman performance on various perfect-information games, such as chess, shogi and Go. However, directly applying AlphaZero to imperfect-information games (IIG) is infeasible, due to the fact that traditional MCTS methods cannot handle missing information of other players. Meanwhile, there have been several extensions of MCTS for IIGs, by implicitly or explicitly sampling a state of other players. But, due to the inability to handle private and public information well, the performance of these methods is not satisfactory. In this paper, we extend AlphaZero to multiplayer IIGs by developing a new MCTS method, Action-Prediction MCTS (AP-MCTS). In contrast to traditional MCTS extensions for IIGs, AP-MCTS first builds the search tree based on public information, adopts the policy-value network to generalize between hidden states, and finally predicts other players' actions directly. This design bypasses the inefficiency of sampling and the difficulty of predicting the state of other players. We conduct extensive experiments on the popular 3-player poker game DouDiZhu to evaluate the performance of AP-MCTS combined with the framework AlphaZero. When playing against experienced human players, AP-MCTS achieved a 65.65\% winning rate, which is almost twice the human's winning rate. When comparing with state-of-the-art DouDiZhu AIs, the Elo rating of AP-MCTS is 50 to 200 higher than them. The ablation study shows that accurate action prediction is the key to AP-MCTS winning.

In Goal-oriented Reinforcement learning, relabeling the raw goals in past experience to provide agents with hindsight ability is a major solution to the reward sparsity problem. In this paper, to enhance the diversity of relabeled goals, we develop FGI (Foresight Goal Inference), a new relabeling strategy that relabels the goals by looking into the future with a learned dynamics model. Besides, to improve sample efficiency, we propose to use the dynamics model to generate simulated trajectories for policy training. By integrating these two improvements, we introduce the MapGo framework (Model-Assisted Policy optimization for Goal-oriented tasks). In our experiments, we first show the effectiveness of the FGI strategy compared with the hindsight one, and then show that the MapGo framework achieves higher sample efficiency when compared to model-free baselines on a set of complicated tasks.

Recently, many researchers have made successful progress in building the AI systems for MOBA-game-playing with deep reinforcement learning, such as on Dota 2 and Honor of Kings. Even though these AI systems have achieved or even exceeded human-level performance, they still suffer from the lack of policy diversity. In this paper, we propose a novel Macro-Goals Guided framework, called MGG, to learn diverse policies in MOBA games. MGG abstracts strategies as macro-goals from human demonstrations and trains a Meta-Controller to predict these macro-goals. To enhance policy diversity, MGG samples macro-goals from the Meta-Controller prediction and guides the training process towards these goals. Experimental results on the typical MOBA game Honor of Kings demonstrate that MGG can execute diverse policies in different matches and lineups, and also outperform the state-of-the-art methods over 102 heroes.

Hero drafting is essential in multiplayer online battle arena (MOBA) game playing as it builds the team of each side and directly affects the match outcome. State-of-the-art drafting methods fail to consider: 1) drafting efficiency when the hero pool is expanded; 2) the multiround nature of a MOBA 5v5 match series, i.e., two teams play best-of-$N$and the same hero is only allowed to be drafted once throughout the series. In this article, we formulate the drafting process as a multiround combinatorial game and propose a novel drafting algorithm based on neural networks and Monte Carlo tree search, named JueWuDraft. Specifically, we design a long-term value estimation mechanism to handle the best-of-$N$drafting case. TakingHonor of Kings, one of the most popular MOBA games at present, as a running case, we demonstrate the practicality and effectiveness of JueWuDraft when compared to state-of-the-art drafting methods.

We study the reinforcement learning problem of complex action control in the Multi-player Online Battle Arena (MOBA) 1v1 games. This problem involves far more complicated state and action spaces than those of traditional 1v1 games, such as Go and Atari series, which makes it very difficult to search any policies with human-level performance. In this paper, we present a deep reinforcement learning framework to tackle this problem from the perspectives of both system and algorithm. Our system is of low coupling and high scalability, which enables efficient explorations at large scale. Our algorithm includes several novel strategies, including control dependency decoupling, action mask, target attention, and dual-clip PPO, with which our proposed actor-critic network can be effectively trained in our system. Tested on the MOBA game Honor of Kings, the trained AI agents can defeat top professional human players in full 1v1 games.

MOBA games, e.g., Honor of Kings, League of Legends, and Dota 2, pose grand challenges to AI systems such as multi-agent, enormous state-action space, complex action control, etc. Developing AI for playing MOBA games has raised much attention accordingly. However, existing work falls short in handling the raw game complexity caused by the explosion of agent combinations, i.e., lineups, when expanding the hero pool in case that OpenAI's Dota AI limits the play to a pool of only 17 heroes. As a result, full MOBA games without restrictions are far from being mastered by any existing AI system. In this paper, we propose a MOBA AI learning paradigm that methodologically enables playing full MOBA games with deep reinforcement learning. Specifically, we develop a combination of novel and existing learning techniques, including off-policy adaption, multi-head value estimation, curriculum self-play learning, policy distillation, and Monte-Carlo tree-search, in training and playing a large pool of heroes, meanwhile addressing the scalability issue skillfully. Tested on Honor of Kings, a popular MOBA game, we show how to build superhuman AI agents that can defeat top esports players. The superiority of our AI is demonstrated by the first large-scale performance test of MOBA AI agent in the literature.