We introduce Mean-Field Trust Region Policy Optimization (MF-TRPO), a novel algorithm designed to compute approximate Nash equilibria for ergodic Mean-Field Games (MFG) in finite state-action spaces. Building on the well-established performance of TRPO in the reinforcement learning (RL) setting, we extend its methodology to the MFG framework, leveraging its stability and robustness in policy optimization. Under standard assumptions in the MFG literature, we provide a rigorous analysis of MF-TRPO, establishing theoretical guarantees on its convergence. Our results cover both the exact formulation of the algorithm and its sample-based counterpart, where we derive high-probability guarantees and finite sample complexity. This work advances MFG optimization by bridging RL techniques with mean-field decision-making, offering a theoretically grounded approach to solving complex multi-agent problems.

Multi-agent reinforcement learning (MARL) methods typically require that agents enjoy global state observability, preventing development of decentralized algorithms and limiting scalability. Recent work has shown that, under assumptions on decaying inter-agent influence, global observability can be replaced by local neighborhood observability at each agent, enabling decentralization and scalability. Real-world applications enjoying such decay properties remain underexplored, however, despite the fact that signal power decay, or signal attenuation, due to path loss is an intrinsic feature of many problems in wireless communications and radar networks. In this paper, we show that signal attenuation enables decentralization in MARL by considering the illustrative special case of performing power allocation for target detection in a radar network. To achieve this, we propose two new constrained multi-agent Markov decision process formulations of this power allocation problem, derive local neighborhood approximations for global value function and policy gradient estimates and establish corresponding error bounds, and develop decentralized saddle point policy gradient algorithms for solving the proposed problems. Our approach, though oriented towards the specific radar network problem we consider, provides a useful model for extensions to additional problems in wireless communications and radar networks.

Reinforcement learning (RL) has significantly advanced the control of physics-based and robotic characters that track kinematic reference motion. However, methods typically rely on a weighted sum of conflicting reward functions, requiring extensive tuning to achieve a desired behavior. Due to the computational cost of RL, this iterative process is a tedious, time-intensive task. Furthermore, for robotics applications, the weights need to be chosen such that the policy performs well in the real world, despite inevitable sim-to-real gaps. To address these challenges, we propose a multi-objective reinforcement learning framework that trains a single policy conditioned on a set of weights, spanning the Pareto front of reward trade-offs. Within this framework, weights can be selected and tuned after training, significantly speeding up iteration time. We demonstrate how this improved workflow can be used to perform highly dynamic motions with a robot character. Moreover, we explore how weight-conditioned policies can be leveraged in hierarchical settings, using a high-level policy to dynamically select weights according to the current task. We show that the multi-objective policy encodes a diverse spectrum of behaviors, facilitating efficient adaptation to novel tasks.

People employ efficient planning strategies. But how are these strategies acquired? Previous research suggests that people can discover new planning strategies through learning from reinforcements, a process known as metacognitive reinforcement learning (MCRL). While prior work has shown that MCRL models can learn new planning strategies and explain more participants' experience-driven discovery better than alternative mechanisms, it also revealed significant individual differences in metacognitive learning. Furthermore, when fitted to human data, these models exhibit a slower rate of strategy discovery than humans. In this study, we investigate whether incorporating cognitive mechanisms that might facilitate human strategy discovery can bring models of MCRL closer to human performance. Specifically, we consider intrinsically generated metacognitive pseudo-rewards, subjective effort valuation, and termination deliberation. Analysis of planning task data shows that a larger proportion of participants used at least one of these mechanisms, with significant individual differences in their usage and varying impacts on strategy discovery. Metacognitive pseudo-rewards, subjective effort valuation, and learning the value of acting without further planning were found to facilitate strategy discovery. While these enhancements provided valuable insights into individual differences and the effect of these mechanisms on strategy discovery, they did not fully close the gap between model and human performance, prompting further exploration of additional factors that people might use to discover new planning strategies.

Ensuring convergence of policy gradient methods in federated reinforcement learning (FRL) under environment heterogeneity remains a major challenge. In this work, we first establish that heterogeneity, perhaps counter-intuitively, can necessitate optimal policies to be non-deterministic or even time-varying, even in tabular environments. Subsequently, we prove global convergence results for federated policy gradient ( FedPG) algorithms employing local updates, under a Łojasiewicz condition that holds only for each individual agent, in both entropy-regularized and non-regularized scenarios. Crucially, our theoretical analysis shows that FedPG attains linear speed-up with respect to the number of agents, a property central to efficient federated learning. Leveraging insights from our theoretical findings, we introduce b-RS-FedPG, a novel policy gradient method that employs a carefully constructed softmax-inspired parameterization coupled with an appropriate regularization scheme. We further demonstrate explicit convergence rates for b-RS-FedPG toward near-optimal stationary policies. Finally, we demonstrate that empirically both FedPG and b-RS-FedPG consistently outperform federated Q-learning on heterogeneous settings.

At the core of reinforcement learning is the idea of learning beyond the performance in the data. However, scaling such systems has proven notoriously tricky. In contrast, techniques from generative modeling have proven remarkably scalable and are simple to train. In this work, we combine these strengths, by deriving a direct relation between policy improvement and guidance of diffusion models. The resulting framework, CFGRL, is trained with the simplicity of supervised learning, yet can further improve on the policies in the data. On offline RL tasks, we observe a reliable trend -- increased guidance weighting leads to increased performance. Of particular importance, CFGRL can operate without explicitly learning a value function, allowing us to generalize simple supervised methods (e.g., goal-conditioned behavioral cloning) to further prioritize optimality, gaining performance for "free" across the board.

Process reward models (PRMs) provide more nuanced supervision compared to outcome reward models (ORMs) for optimizing policy models, positioning them as a promising approach to enhancing the capabilities of LLMs in complex reasoning tasks. Recent efforts have advanced PRMs from step-level to token-level granularity by integrating reward modeling into the training of generative models, with reward scores derived from token generation probabilities. However, the conflict between generative language modeling and reward modeling may introduce instability and lead to inaccurate credit assignments. To address this challenge, we revisit token-level reward assignment by decoupling reward modeling from language generation and derive a token-level reward model through the optimization of a discriminative policy, termed the Q-function Reward Model (Q-RM). We theoretically demonstrate that Q-RM explicitly learns token-level Q-functions from preference data without relying on fine-grained annotations. In our experiments, Q-RM consistently outperforms all baseline methods across various benchmarks. For example, when integrated into PPO/REINFORCE algorithms, Q-RM enhances the average Pass@1 score by 5.85/4.70 points on mathematical reasoning tasks compared to the ORM baseline, and by 4.56/5.73 points compared to the token-level PRM counterpart. Moreover, reinforcement learning with Q-RM significantly enhances training efficiency, achieving convergence 12 times faster than ORM on GSM8K and 11 times faster than step-level PRM on MATH. Code and data are available at https://github.com/homzer/Q-RM.

Recent advances in language modeling and vision stem from training large models on diverse, multi-task data. This paradigm has had limited impact in value-based reinforcement learning (RL), where improvements are often driven by small models trained in a single-task context. This is because in multi-task RL sparse rewards and gradient conflicts make optimization of temporal difference brittle. Practical workflows for generalist policies therefore avoid online training, instead cloning expert trajectories or distilling collections of single-task policies into one agent. In this work, we show that the use of high-capacity value models trained via cross-entropy and conditioned on learnable task embeddings addresses the problem of task interference in online RL, allowing for robust and scalable multi-task training. We test our approach on 7 multi-task benchmarks with over 280 unique tasks, spanning high degree-of-freedom humanoid control and discrete vision-based RL. We find that, despite its simplicity, the proposed approach leads to state-of-the-art single and multi-task performance, as well as sample-efficient transfer to new tasks.

Acknowledgements: The work was supported by the National Natural Science Foundation of China [Grants 72471216, 72022018, 72091210] and Youth Innovation Promotion Association, Chinese Academy of Sciences [Grant No. 2021454]. A specialized fitness metric quantifying customers' temporal flexibility enhances the shaking phase effectiveness. Computational experiments on realistic unmanned vending machine replenishment scenarios demonstrate RL-AVNS's superior performance. The approach exhibits strong generalization capabilities to unseen problem instances, offering practical value for complex logistics optimization. Learning to Search for Vehicle Routing with Multiple Time Windows A R T I C L E I N F OKeywords: Vehicle routing Multiple time windows Reinforcement learning Unmanned vending machine replenishment A B S T R A C T In this study, we propose a reinforcement learning-based adaptive variable neighborhood search (RL-AVNS) method designed for effectively solving the Vehicle Routing Problem with Multiple Time Windows (VRPMTW). Unlike traditional adaptive approaches that rely solely on historical operator performance, our method integrates a reinforcement learning framework to dynamically select neighborhood operators based on real-time solution states and learned experience. We introduce a fitness metric that quantifies customers' temporal flexibility to improve the shaking phase, and employ a transformer-based neural policy network to intelligently guide operator selection during the local search. Extensive computational experiments are conducted on realistic scenarios derived from the replenishment of unmanned vending machines, characterized by multiple clustered replenishment windows. Results demonstrate that RL-AVNS significantly outperforms traditional variable neighborhood search (VNS), adaptive VNS (AVNS), and state-of-the-art learning-based heuristics, achieving substantial improvements in solution quality and computational efficiency across various instance scales and time window complexities. Particularly notable is the algorithm's capability to generalize effectively to problem instances not encountered during training, underscoring its practical utility for complex logistics scenarios.1. Introduction Vehicle Routing Problems (VRPs) are fundamental to optimizing logistics and transportation systems. They are critical for ensuring timely and cost-effective deliveries in various industries, including e-commerce, healthcare, and food services (Vigo and Toth, 2014; Cordeau et al., 2002). In response to growing customer expectations for personalized services, logistics providers are increasingly offering flexible delivery options to improve service quality and maintain a competitive edge.

Robust reinforcement learning (RL) aims to learn policies that remain effective despite uncertainties in its environment, which frequently arise in real-world applications due to variations in environment dynamics. The robust RL methods learn a robust policy by maximizing value under the worst-case models within a predefined uncertainty set. Offline robust RL algorithms are particularly promising in scenarios where only a fixed dataset is available and new data cannot be collected. However, these approaches often require extensive offline data, and gathering such datasets for specific tasks in specific environments can be both costly and time-consuming. Using an imperfect simulator offers a faster, cheaper, and safer way to collect data for training, but it can suffer from dynamics mismatch. In this paper, we introduce HYDRO, the first Hybrid Cross-Domain Robust RL framework designed to address these challenges. HYDRO utilizes an online simulator to complement the limited amount of offline datasets in the non-trivial context of robust RL. By measuring and minimizing performance gaps between the simulator and the worst-case models in the uncertainty set, HYDRO employs novel uncertainty filtering and prioritized sampling to select the most relevant and reliable simulator samples. Our extensive experiments demonstrate HYDRO's superior performance over existing methods across various tasks, underscoring its potential to improve sample efficiency in offline robust RL.