ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates

We present that hierarchical LLM reasoning via scaling thought templates can effectively optimize the reasoning search space and outperform the mathematical reasoning capabilities of powerful LLMs like OpenAI o1-preview and DeepSeek V3. We train our ReasonFlux-32B model with only 8 GPUs and introduces three innovations: (i) a structured and generic thought template library, containing around 500 high-level thought templates capable of generalizing to similar or relevant reasoning problems; (ii) performing hierarchical reinforcement learning on a sequence of thought templates instead of long CoTs, optimizing a base LLM to plan out an optimal template trajectory for gradually handling complex problems; (iii) a brand new inference scaling system that enables hierarchical LLM reasoning by adaptively scaling thought templates at inference time. With a template trajectory containing sequential thought templates, our ReasonFlux-32B significantly advances math reasoning capabilities to state-of-the-art levels. Notably, on the MATH benchmark, it achieves an accuracy of 91.2% and surpasses o1-preview by 6.7%. On the USA Math Olympiad (AIME) benchmark, ReasonFlux-32B solves an average of 56.7% of problems, surpassing o1-preview and DeepSeek-V3 by 27% and 45%, respectively. Code: https://github.com/Gen-Verse/ReasonFlux

Discovery of skill switching criteria for learning agile quadruped locomotion

This paper develops a hierarchical learning and optimization framework that can learn and achieve well-coordinated multi-skill locomotion. The learned multi-skill policy can switch between skills automatically and naturally in tracking arbitrarily positioned goals and recover from failures promptly. The proposed framework is composed of a deep reinforcement learning process and an optimization process. First, the contact pattern is incorporated into the reward terms for learning different types of gaits as separate policies without the need for any other references. Then, a higher level policy is learned to generate weights for individual policies to compose multi-skill locomotion in a goal-tracking task setting. Skills are automatically and naturally switched according to the distance to the goal. The proper distances for skill switching are incorporated in reward calculation for learning the high level policy and updated by an outer optimization loop as learning progresses. We first demonstrated successful multi-skill locomotion in comprehensive tasks on a simulated Unitree A1 quadruped robot. We also deployed the learned policy in the real world showcasing trotting, bounding, galloping, and their natural transitions as the goal position changes. Moreover, the learned policy can react to unexpected failures at any time, perform prompt recovery, and resume locomotion successfully. Compared to discrete switch between single skills which failed to transition to galloping in the real world, our proposed approach achieves all the learned agile skills, with smoother and more continuous skill transitions.

Sequential Stochastic Combinatorial Optimization Using Hierarchal Reinforcement Learning

Reinforcement learning (RL) has emerged as a promising tool for combinatorial optimization (CO) problems due to its ability to learn fast, effective, and generalizable solutions. Nonetheless, existing works mostly focus on one-shot deterministic CO, while sequential stochastic CO (SSCO) has rarely been studied despite its broad applications such as adaptive influence maximization (IM) and infectious disease intervention. In this paper, we study the SSCO problem where we first decide the budget (e.g., number of seed nodes in adaptive IM) allocation for all time steps, and then select a set of nodes for each time step. The few existing studies on SSCO simplify the problems by assuming a uniformly distributed budget allocation over the time horizon, yielding suboptimal solutions. We propose a generic hierarchical RL (HRL) framework called wake-sleep option (WS-option), a two-layer option-based framework that simultaneously decides adaptive budget allocation on the higher layer and node selection on the lower layer. WS-option starts with a coherent formulation of the two-layer Markov decision processes (MDPs), capturing the interdependencies between the two layers of decisions. Building on this, WS-option employs several innovative designs to balance the model's training stability and computational efficiency, preventing the vicious cyclic interference issue between the two layers. Empirical results show that WS-option exhibits significantly improved effectiveness and generalizability compared to traditional methods. Moreover, the learned model can be generalized to larger graphs, which significantly reduces the overhead of computational resources.

LLM-Powered Decentralized Generative Agents with Adaptive Hierarchical Knowledge Graph for Cooperative Planning

Developing intelligent agents for long-term cooperation in dynamic open-world scenarios is a major challenge in multi-agent systems. Traditional Multi-agent Reinforcement Learning (MARL) frameworks like centralized training decentralized execution (CTDE) struggle with scalability and flexibility. They require centralized long-term planning, which is difficult without custom reward functions, and face challenges in processing multi-modal data. CTDE approaches also assume fixed cooperation strategies, making them impractical in dynamic environments where agents need to adapt and plan independently. To address decentralized multi-agent cooperation, we propose Decentralized Adaptive Knowledge Graph Memory and Structured Communication System (DAMCS) in a novel Multi-agent Crafter environment. Our generative agents, powered by Large Language Models (LLMs), are more scalable than traditional MARL agents by leveraging external knowledge and language for long-term planning and reasoning. Instead of fully sharing information from all past experiences, DAMCS introduces a multi-modal memory system organized as a hierarchical knowledge graph and a structured communication protocol to optimize agent cooperation. This allows agents to reason from past interactions and share relevant information efficiently. Experiments on novel multi-agent open-world tasks show that DAMCS outperforms both MARL and LLM baselines in task efficiency and collaboration. Compared to single-agent scenarios, the two-agent scenario achieves the same goal with 63% fewer steps, and the six-agent scenario with 74% fewer steps, highlighting the importance of adaptive memory and structured communication in achieving long-term goals. We publicly release our project at: https://happyeureka.github.io/damcs.

Bilevel Multi-Armed Bandit-Based Hierarchical Reinforcement Learning for Interaction-Aware Self-Driving at Unsignalized Intersections

In this work, we present BiM-ACPPO, a bilevel multi-armed bandit-based hierarchical reinforcement learning framework for interaction-aware decision-making and planning at unsignalized intersections. Essentially, it proactively takes the uncertainties associated with surrounding vehicles (SVs) into consideration, which encompass those stemming from the driver's intention, interactive behaviors, and the varying number of SVs. Intermediate decision variables are introduced to enable the high-level RL policy to provide an interaction-aware reference, for guiding low-level model predictive control (MPC) and further enhancing the generalization ability of the proposed framework. By leveraging the structured nature of self-driving at unsignalized intersections, the training problem of the RL policy is modeled as a bilevel curriculum learning task, which is addressed by the proposed Exp3.S-based BiMAB algorithm. It is noteworthy that the training curricula are dynamically adjusted, thereby facilitating the sample efficiency of the RL training process. Comparative experiments are conducted in the high-fidelity CARLA simulator, and the results indicate that our approach achieves superior performance compared to all baseline methods. Furthermore, experimental results in two new urban driving scenarios clearly demonstrate the commendable generalization performance of the proposed method.

Hierarchical Contextual Manifold Alignment for Structuring Latent Representations in Large Language Models

The organization of latent token representations plays a crucial role in determining the stability, generalization, and contextual consistency of language models, yet conventional approaches to embedding refinement often rely on parameter modifications that introduce additional computational overhead. A hierarchical alignment method was introduced to restructure token embeddings without altering core model weights, ensuring that representational distributions maintained coherence across different linguistic contexts. Experimental evaluations demonstrated improvements in rare token retrieval, adversarial robustness, and long-range dependency tracking, highlighting the advantages of hierarchical structuring in mitigating inconsistencies in latent space organization. The comparative analysis against conventional fine-tuning and embedding perturbation methods revealed that hierarchical restructuring maintained computational efficiency while achieving measurable gains in representation quality. Structural refinements introduced through the alignment process resulted in improved contextual stability across varied linguistic tasks, reducing inconsistencies in token proximity relationships and enhancing interpretability in language generation. A detailed computational assessment confirmed that the realignment process introduced minimal inference overhead, ensuring that representational improvements did not compromise model efficiency. The findings reinforced the broader significance of structured representation learning, illustrating that hierarchical embedding modifications could serve as an effective strategy for refining latent space distributions while preserving pre-learned semantic associations.

DHP: Discrete Hierarchical Planning for Hierarchical Reinforcement Learning Agents

In this paper, we address the challenge of long-horizon visual planning tasks using Hierarchical Reinforcement Learning (HRL). Our key contribution is a Discrete Hierarchical Planning (DHP) method, an alternative to traditional distance-based approaches. We provide theoretical foundations for the method and demonstrate its effectiveness through extensive empirical evaluations. Our agent recursively predicts subgoals in the context of a long-term goal and receives discrete rewards for constructing plans as compositions of abstract actions. The method introduces a novel advantage estimation strategy for tree trajectories, which inherently encourages shorter plans and enables generalization beyond the maximum tree depth. The learned policy function allows the agent to plan efficiently, requiring only $\log N$ computational steps, making re-planning highly efficient. The agent, based on a soft-actor critic (SAC) framework, is trained using on-policy imagination data. Additionally, we propose a novel exploration strategy that enables the agent to generate relevant training examples for the planning modules. We evaluate our method on long-horizon visual planning tasks in a 25-room environment, where it significantly outperforms previous benchmarks at success rate and average episode length. Furthermore, an ablation study highlights the individual contributions of key modules to the overall performance.

CH-MARL: Constrained Hierarchical Multiagent Reinforcement Learning for Sustainable Maritime Logistics

Addressing global challenges such as greenhouse gas emissions and resource inequity demands advanced AI-driven coordination among autonomous agents. We propose CH-MARL (Constrained Hierarchical Multiagent Reinforcement Learning), a novel framework that integrates hierarchical decision-making with dynamic constraint enforcement and fairness-aware reward shaping. CH-MARL employs a real-time constraint-enforcement layer to ensure adherence to global emission caps, while incorporating fairness metrics that promote equitable resource distribution among agents. Experiments conducted in a simulated maritime logistics environment demonstrate considerable reductions in emissions, along with improvements in fairness and operational efficiency. Beyond this domain-specific success, CH-MARL provides a scalable, generalizable solution to multi-agent coordination challenges in constrained, dynamic settings, thus advancing the state of the art in reinforcement learning.

PRISM: A Robust Framework for Skill-based Meta-Reinforcement Learning with Noisy Demonstrations

Meta-reinforcement learning (Meta-RL) facilitates rapid adaptation to unseen tasks but faces challenges in long-horizon environments. Skill-based approaches tackle this by decomposing state-action sequences into reusable skills and employing hierarchical decision-making. However, these methods are highly susceptible to noisy offline demonstrations, resulting in unstable skill learning and degraded performance. To overcome this, we propose Prioritized Refinement for Skill-Based Meta-RL (PRISM), a robust framework that integrates exploration near noisy data to generate online trajectories and combines them with offline data. Through prioritization, PRISM extracts high-quality data to learn task-relevant skills effectively. By addressing the impact of noise, our method ensures stable skill learning and achieves superior performance in long-horizon tasks, even with noisy and sub-optimal data.

Hierarchical Consensus Network for Multiview Feature Learning

Multiview feature learning aims to learn discriminative features by integrating the distinct information in each view. However, most existing methods still face significant challenges in learning view-consistency features, which are crucial for effective multiview learning. Motivated by the theories of CCA and contrastive learning in multiview feature learning, we propose the hierarchical consensus network (HCN) in this paper. The HCN derives three consensus indices for capturing the hierarchical consensus across views, which are classifying consensus, coding consensus, and global consensus, respectively. Specifically, classifying consensus reinforces class-level correspondence between views from a CCA perspective, while coding consensus closely resembles contrastive learning and reflects contrastive comparison of individual instances. Global consensus aims to extract consensus information from two perspectives simultaneously. By enforcing the hierarchical consensus, the information within each view is better integrated to obtain more comprehensive and discriminative features. The extensive experimental results obtained on four multiview datasets demonstrate that the proposed method significantly outperforms several state-of-the-art methods.