Building Behavioral Foundation Models (BFMs) for humanoid robots has the potential to unify diverse control tasks under a single, promptable generalist policy. However, existing approaches are either exclusively deployed on simulated humanoid characters, or specialized to specific tasks such as tracking. We propose BFM-Zero, a framework that learns an effective shared latent representation that embeds motions, goals, and rewards into a common space, enabling a single policy to be prompted for multiple downstream tasks without retraining. This well-structured latent space in BFM-Zero enables versatile and robust whole-body skills on a Unitree G1 humanoid in the real world, via diverse inference methods, including zero-shot motion tracking, goal reaching, and reward optimization, and few-shot optimization-based adaptation. Unlike prior on-policy reinforcement learning (RL) frameworks, BFM-Zero builds upon recent advancements in unsupervised RL and Forward-Backward (FB) models, which offer an objective-centric, explainable, and smooth latent representation of whole-body motions. We further extend BFM-Zero with critical reward shaping, domain randomization, and history-dependent asymmetric learning to bridge the sim-to-real gap. Those key design choices are quantitatively ablated in simulation. A first-of-its-kind model, BFM-Zero establishes a step toward scalable, promptable behavioral foundation models for whole-body humanoid control.

Humanoid soccer poses a representative challenge for embodied intelligence, requiring robots to operate within a tightly coupled perception-action loop. However, existing systems typically rely on decoupled modules, resulting in delayed responses and incoherent behaviors in dynamic environments, while real-world perceptual limitations further exacerbate these issues. In this work, we present a unified reinforcement learning-based controller that enables humanoid robots to acquire reactive soccer skills through the direct integration of visual perception and motion control. Our approach extends Adversarial Motion Priors to perceptual settings in real-world dynamic environments, bridging motion imitation and visually grounded dynamic control. We introduce an encoder-decoder architecture combined with a virtual perception system that models real-world visual characteristics, allowing the policy to recover privileged states from imperfect observations and establish active coordination between perception and action. The resulting controller demonstrates strong reactivity, consistently executing coherent and robust soccer behaviors across various scenarios, including real RoboCup matches.

In multi-agent deep reinforcement learning (MADRL), agents can communicate with one another to perform a task in a coordinated manner. When multiple tasks are involved, agents can also leverage knowledge from one task to improve learning in other tasks. In this paper, we propose Multi-task Communication Skills (MCS), a MADRL with communication method that learns and performs multiple tasks simultaneously, with agents interacting through learnable communication protocols. MCS employs a Transformer encoder to encode task-specific observations into a shared message space, capturing shared communication skills among agents. To enhance coordination among agents, we introduce a prediction network that correlates messages with the actions of sender agents in each task. We adapt three multi-agent benchmark environments to multi-task settings, where the number of agents as well as the observation and action spaces vary across tasks. Experimental results demonstrate that MCS achieves better performance than multi-task MADRL baselines without communication, as well as single-task MADRL baselines with and without communication.

Large language models (LLMs) have significantly advanced in reasoning tasks through reinforcement learning (RL) optimization, achieving impressive capabilities across various challenging benchmarks. However, our empirical analysis reveals a critical drawback: reasoning-oriented RL fine-tuning significantly increases the prevalence of hallucinations. We theoretically analyze the RL training dynamics, identifying high-variance gradient, entropy-induced randomness, and susceptibility to spurious local optima as key factors leading to hallucinations. To address this drawback, we propose Factuality-aware Step-wise Policy Optimization (FSPO), an innovative RL fine-tuning algorithm incorporating explicit factuality verification at each reasoning step. FSPO leverages automated verification against given evidence to dynamically adjust token-level advantage values, incentivizing factual correctness throughout the reasoning process. Experiments across mathematical reasoning and hallucination benchmarks using Qwen2.5 and Llama models demonstrate that FSPO effectively reduces hallucinations while enhancing reasoning accuracy, substantially improving both reliability and performance.

Deep Q-Networks (DQNs) estimate future returns by learning from transitions sampled from a replay buffer. However, the target updates in DQN often rely on next states generated by actions from past, potentially suboptimal, policy. As a result, these states may not provide informative learning signals, causing high variance into the update process. This issue is exacerbated when the sampled transitions are poorly aligned with the agent's current policy. To address this limitation, we propose the Successor-state Aggregation Deep Q-Network (SADQ), which explicitly models environment dynamics using a stochastic transition model. SADQ integrates successor-state distributions into the Q-value estimation process, enabling more stable and policy-aligned value updates. Additionally, it explores a more efficient action selection strategy with the modeled transition structure. We provide theoretical guarantees that SADQ maintains unbiased value estimates while reducing training variance. Our extensive empirical results across standard RL benchmarks and real-world vector-based control tasks demonstrate that SADQ consistently outperforms DQN variants in both stability and learning efficiency.

Entropy Minimization (EM) is beneficial to reducing class overlap, bridging domain gap, and restricting uncertainty for various tasks in machine learning, yet its potential is limited. To study the internal mechanism of EM, we reformulate and decouple the classical EM into two parts with opposite effects: cluster aggregation driving factor (CADF) rewards dominant classes and prompts a peaked output distribution, while gradient mitigation calibrator (GMC) penalizes high-confidence classes based on predicted probabilities. Furthermore, we reveal the limitations of classical EM caused by its coupled formulation: 1) reward collapse impedes the contribution of high-certainty samples in the learning process, and 2) easy-class bias induces misalignment between output distribution and label distribution. To address these issues, we propose Adaptive Decoupled Entropy Minimization (AdaDEM), which normalizes the reward brought from CADF and employs a marginal entropy calibrator (MEC) to replace GMC. AdaDEM outperforms DEM*, an upper-bound variant of classical EM, and achieves superior performance across various imperfectly supervised learning tasks in noisy and dynamic environments.

Low-rank structure is a common implicit assumption in many modern reinforcement learning (RL) algorithms. For instance, reward-free and goal-conditioned RL methods often presume that the successor measure admits a low-rank representation. In this work, we challenge this assumption by first remarking that the successor measure itself is not approximately low-rank. Instead, we demonstrate that a low-rank structure naturally emerges in the shifted successor measure, which captures the system dynamics after bypassing a few initial transitions. We provide finite-sample performance guarantees for the entry-wise estimation of a low-rank approximation of the shifted successor measure from sampled entries. Our analysis reveals that both the approximation and estimation errors are primarily governed by a newly introduced quantitity: the spectral recoverability of the corresponding matrix. To bound this parameter, we derive a new class of functional inequalities for Markov chains that we call Type II Poincaré inequalities and from which we can quantify the amount of shift needed for effective low-rank approximation and estimation. This analysis shows in particular that the required shift depends on decay of the high-order singular values of the shifted successor measure and is hence typically small in practice. Additionally, we establish a connection between the necessary shift and the local mixing properties of the underlying dynamical system, which provides a natural way of selecting the shift. Finally, we validate our theoretical findings with experiments, and demonstrate that shifting the successor measure indeed leads to improved performance in goal-conditioned RL.

As cloud computing and microservice architectures become increasingly prevalent, API rate limiting has emerged as a critical mechanism for ensuring system stability and service quality. Traditional rate limiting algorithms, such as token bucket and sliding window, while widely adopted, struggle to adapt to dynamic traffic patterns and varying system loads. This paper proposes an adaptive rate limiting strategy based on deep reinforcement learning that dynamically balances system throughput and service latency. We design a hybrid architecture combining Deep Q-Network (DQN) and Asynchronous Advantage Actor-Critic (A3C) algorithms, modeling the rate limiting decision process as a Markov Decision Process. The system continuously monitors microservice states and learns optimal rate limiting policies through environmental interaction. Extensive experiments conducted in a Kubernetes cluster environment demonstrate that our approach achieves 23.7% throughput improvement and 31.4% P99 latency reduction compared to traditional fixed-threshold strategies under high-load scenarios. Results from a 90-day production deployment handling 500 million daily requests validate the practical effectiveness of the proposed method, with 82% reduction in service degradation incidents and 68% decrease in manual interventions.

Abstract--The rapid expansion of immersive Metaverse applications introduces complex challenges at the intersection of performance, privacy, and environmental sustainability. Centralized architectures fall short in addressing these demands, often resulting in elevated energy consumption, latency, and privacy concerns. This paper proposes MetaF ed, a decentralized federated learning (FL) framework that enables sustainable and intelligent resource orchestration for Metaverse environments. MetaFed integrates (i) multi-agent reinforcement learning for dynamic client selection, (ii) privacy-preserving FL using homomorphic encryption, and (iii) carbon-aware scheduling aligned with renewable energy availability. Evaluations on MNIST and CIF AR-10 using lightweight ResNet architectures demonstrate that MetaFed achieves up to 25% reduction in carbon emissions compared to conventional approaches, while maintaining high accuracy and minimal communication overhead.

The exponential growth of Low Earth Orbit (LEO) satellites has revolutionised Earth Observation (EO) missions, addressing challenges in climate monitoring, disaster management, and more. However, autonomous coordination in multi-satellite systems remains a fundamental challenge. Traditional optimisation approaches struggle to handle the real-time decision-making demands of dynamic EO missions, necessitating the use of Reinforcement Learning (RL) and Multi-Agent Reinforcement Learning (MARL). In this paper, we investigate RL-based autonomous EO mission planning by modelling single-satellite operations and extending to multi-satellite constellations using MARL frameworks. We address key challenges, including energy and data storage limitations, uncertainties in satellite observations, and the complexities of decentralised coordination under partial observability. By leveraging a near-realistic satellite simulation environment, we evaluate the training stability and performance of state-of-the-art MARL algorithms, including PPO, IPPO, MAPPO, and HAPPO. Our results demonstrate that MARL can effectively balance imaging and resource management while addressing non-stationarity and reward interdependency in multi-satellite coordination. The insights gained from this study provide a foundation for autonomous satellite operations, offering practical guidelines for improving policy learning in decentralised EO missions.