Humanoid robots hold significant potential in accomplishing daily tasks across diverse environments thanks to their flexibility and human-like morphology. Recent works have made significant progress in humanoid whole-body control and loco-manipulation leveraging optimal control or reinforcement learning. However, these methods require tedious task-specific tuning for each task to achieve satisfactory behaviors, limiting their versatility and scalability to diverse tasks in daily scenarios. To that end, we introduce SkillBlender, a novel hierarchical reinforcement learning framework for versatile humanoid loco-manipulation. SkillBlender first pretrains goal-conditioned task-agnostic primitive skills, and then dynamically blends these skills to accomplish complex loco-manipulation tasks with minimal task-specific reward engineering. We also introduce SkillBench, a parallel, cross-embodiment, and diverse simulated benchmark containing three embodiments, four primitive skills, and eight challenging loco-manipulation tasks, accompanied by a set of scientific evaluation metrics balancing accuracy and feasibility. Extensive simulated experiments show that our method significantly outperforms all baselines, while naturally regularizing behaviors to avoid reward hacking, resulting in more accurate and feasible movements for diverse loco-manipulation tasks in our daily scenarios. Our code and benchmark will be open-sourced to the community to facilitate future research. Project page: https://usc-gvl.github.io/SkillBlender-web/.
Multi-agent frameworks can substantially boost the reasoning power of large language models (LLMs), but they typically incur heavy computational costs and lack convergence guarantees. To overcome these challenges, we recast multi-LLM coordination as an incomplete-information game and seek a Bayesian Nash equilibrium (BNE), in which each agent optimally responds to its probabilistic beliefs about the strategies of others. We introduce Efficient Coordination via Nash Equilibrium (ECON), a hierarchical reinforcement-learning paradigm that marries distributed reasoning with centralized final output. Under ECON, each LLM independently selects responses that maximize its expected reward, conditioned on its beliefs about co-agents, without requiring costly inter-agent exchanges. We mathematically prove that ECON attains a markedly tighter regret bound than non-equilibrium multi-agent schemes. Empirically, ECON outperforms existing multi-LLM approaches by 11.2% on average across six benchmarks spanning complex reasoning and planning tasks. Further experiments demonstrate ECON's ability to flexibly incorporate additional models, confirming its scalability and paving the way toward larger, more powerful multi-LLM ensembles. The code is publicly available at: https://github.com/tmlr-group/ECON.
While showing sophisticated reasoning abilities, large language models (LLMs) still struggle with long-horizon decision-making tasks due to deficient exploration and long-term credit assignment, especially in sparse-reward scenarios. Inspired by the divide-and-conquer principle, we propose an innovative framework **GLIDER** (**G**rounding **L**anguage Models as Eff**I**cient **D**ecision-Making Agents via Offline Hi**E**rarchical **R**einforcement Learning) that introduces a parameter-efficient and generally applicable hierarchy to LLM policies. We develop a scheme where the low-level controller is supervised with abstract, step-by-step plans that are learned and instructed by the high-level policy. This design decomposes complicated problems into a series of coherent chain-of-thought reasoning sub-tasks, providing flexible temporal abstraction to significantly enhance exploration and learning for long-horizon tasks. Furthermore, GLIDER facilitates fast online adaptation to non-stationary environments owing to the strong transferability of its task-agnostic low-level skills. Experiments on ScienceWorld and ALFWorld benchmarks show that GLIDER achieves consistent performance gains, along with enhanced generalization capabilities.
In collaborative tasks, autonomous agents fall short of humans in their capability to quickly adapt to new and unfamiliar teammates. We posit that a limiting factor for zero-shot coordination is the lack of shared task abstractions, a mechanism humans rely on to implicitly align with teammates. To address this gap, we introduce HA$^2$: Hierarchical Ad Hoc Agents, a framework leveraging hierarchical reinforcement learning to mimic the structured approach humans use in collaboration. We evaluate HA$^2$ in the Overcooked environment, demonstrating statistically significant improvement over existing baselines when paired with both unseen agents and humans, providing better resilience to environmental shifts, and outperforming all state-of-the-art methods.
We introduce a novel hierarchical reinforcement learning (HRL) framework that performs top-down recursive planning via learned subgoals, successfully applied to the complex combinatorial puzzle game Sokoban. Our approach constructs a six-level policy hierarchy, where each higher-level policy generates subgoals for the level below. All subgoals and policies are learned end-to-end from scratch, without any domain knowledge. Our results show that the agent can generate long action sequences from a single high-level call. While prior work has explored 2-3 level hierarchies and subgoal-based planning heuristics, we demonstrate that deep recursive goal decomposition can emerge purely from learning, and that such hierarchies can scale effectively to hard puzzle domains.
Decision-making in complex, continuous multi-task environments is often hindered by the difficulty of obtaining accurate models for planning and the inefficiency of learning purely from trial and error. While precise environment dynamics may be hard to specify, human experts can often provide high-fidelity abstractions that capture the essential high-level structure of a task and user preferences in the target environment. Existing hierarchical approaches often target discrete settings and do not generalize across tasks. We propose a hierarchical reinforcement learning approach that addresses these limitations by dynamically planning over the expert-specified abstraction to generate subgoals to learn a goal-conditioned policy. To overcome the challenges of learning under sparse rewards, we shape the reward based on the optimal state value in the abstract model. This structured decision-making process enhances sample efficiency and facilitates zero-shot generalization. Our empirical evaluation on a suite of procedurally generated continuous control environments demonstrates that our approach outperforms existing hierarchical reinforcement learning methods in terms of sample efficiency, task completion rate, scalability to complex tasks, and generalization to novel scenarios.
We address the problem of code generation from multi-turn execution feedback. Existing methods either generate code without feedback or use complex, hierarchical reinforcement learning to optimize multi-turn rewards. We propose a simple yet scalable approach, $\mu$Code, that solves multi-turn code generation using only single-step rewards. Our key insight is that code generation is a one-step recoverable MDP, where the correct code can be recovered from any intermediate code state in a single turn. $\mu$Code iteratively trains both a generator to provide code solutions conditioned on multi-turn execution feedback and a verifier to score the newly generated code. Experimental evaluations show that our approach achieves significant improvements over the state-of-the-art baselines. We provide analysis of the design choices of the reward models and policy, and show the efficacy of $\mu$Code at utilizing the execution feedback. Our code is available at https://github.com/portal-cornell/muCode.
Hierarchical organization is fundamental to biological systems and human societies, yet artificial intelligence systems often rely on monolithic architectures that limit adaptability and scalability. Current hierarchical reinforcement learning (HRL) approaches typically restrict hierarchies to two levels or require centralized training, which limits their practical applicability. We introduce TAME Agent Framework (TAG), a framework for constructing fully decentralized hierarchical multi-agent systems. TAG enables hierarchies of arbitrary depth through a novel LevelEnv concept, which abstracts each hierarchy level as the environment for the agents above it. This approach standardizes information flow between levels while preserving loose coupling, allowing for seamless integration of diverse agent types. We demonstrate the effectiveness of TAG by implementing hierarchical architectures that combine different RL agents across multiple levels, achieving improved performance over classical multi-agent RL baselines on standard benchmarks. Our results show that decentralized hierarchical organization enhances both learning speed and final performance, positioning TAG as a promising direction for scalable multi-agent systems.
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
Abstraction is key to scaling up reinforcement learning (RL). However, autonomously learning abstract state and action representations to enable transfer and generalization remains a challenging open problem. This paper presents a novel approach for inventing, representing, and utilizing options, which represent temporally extended behaviors, in continual RL settings. Our approach addresses streams of stochastic problems characterized by long horizons, sparse rewards, and unknown transition and reward functions. Our approach continually learns and maintains an interpretable state abstraction, and uses it to invent high-level options with abstract symbolic representations. These options meet three key desiderata: (1) composability for solving tasks effectively with lookahead planning, (2) reusability across problem instances for minimizing the need for relearning, and (3) mutual independence for reducing interference among options. Our main contributions are approaches for continually learning transferable, generalizable options with symbolic representations, and for integrating search techniques with RL to efficiently plan over these learned options to solve new problems. Empirical results demonstrate that the resulting approach effectively learns and transfers abstract knowledge across problem instances, achieving superior sample efficiency compared to state-of-the-art methods.
Goal-conditioned hierarchical reinforcement learning (HRL) decomposes complex reaching tasks into a sequence of simple subgoal-conditioned tasks, showing significant promise for addressing long-horizon planning in large-scale environments. This paper bridges the goal-conditioned HRL based on graph-based planning to brain mechanisms, proposing a hippocampus-striatum-like dual-controller hypothesis. Inspired by the brain mechanisms of organisms (i.e., the high-reward preferences observed in hippocampal replay) and instance-based theory, we propose a high-return sampling strategy for constructing memory graphs, improving sample efficiency. Additionally, we derive a model-free lower-level Q-function gradient penalty to resolve the model dependency issues present in prior work, improving the generalization of Lipschitz constraints in applications. Finally, we integrate these two extensions, High-reward Graph and model-free Gradient Penalty (HG2P), into the state-of-the-art framework ACLG, proposing a novel goal-conditioned HRL framework, HG2P+ACLG. Experimentally, the results demonstrate that our method outperforms state-of-the-art goal-conditioned HRL algorithms on a variety of long-horizon navigation tasks and robotic manipulation tasks.
Hierarchical Reinforcement Learning (HRL) effectively tackles complex tasks by decomposing them into structured policies. However, HRL agents often face challenges with efficient exploration and rapid adaptation. To address this, we integrate meta-learning into HRL to enhance the agent's ability to learn and adapt hierarchical policies swiftly. Our approach employs meta-learning for rapid task adaptation based on prior experience, while intrinsic motivation mechanisms encourage efficient exploration by rewarding novel state visits. Specifically, our agent uses a high-level policy to select among multiple low-level policies operating within custom grid environments. We utilize gradient-based meta-learning with differentiable inner-loop updates, enabling optimization across a curriculum of increasingly difficult tasks. Experimental results demonstrate that our meta-learned hierarchical agent significantly outperforms traditional HRL agents without meta-learning and intrinsic motivation. The agent exhibits accelerated learning, higher cumulative rewards, and improved success rates in complex grid environments. These findings suggest that integrating meta-learning with HRL, alongside curriculum learning and intrinsic motivation, substantially enhances the agent's capability to handle complex tasks.
In recent years, robots and autonomous systems have become increasingly integral to our daily lives, offering solutions to complex problems across various domains. Their application in search and rescue (SAR) operations, however, presents unique challenges. Comprehensively exploring the disaster-stricken area is often infeasible due to the vastness of the terrain, transformed environment, and the time constraints involved. Traditional robotic systems typically operate on predefined search patterns and lack the ability to incorporate and exploit ground truths provided by human stakeholders, which can be the key to speeding up the learning process and enhancing triage. Addressing this gap, we introduce a system that integrates social interaction via large language models (LLMs) with a hierarchical reinforcement learning (HRL) framework. The proposed system is designed to translate verbal inputs from human stakeholders into actionable RL insights and adjust its search strategy. By leveraging human-provided information through LLMs and structuring task execution through HRL, our approach not only bridges the gap between autonomous capabilities and human intelligence but also significantly improves the agent's learning efficiency and decision-making process in environments characterised by long horizons and sparse rewards.
Recent years, multi-hop reasoning has been widely studied for knowledge graph (KG) reasoning due to its efficacy and interpretability. However, previous multi-hop reasoning approaches are subject to two primary shortcomings. First, agents struggle to learn effective and robust policies at the early phase due to sparse rewards. Second, these approaches often falter on specific datasets like sparse knowledge graphs, where agents are required to traverse lengthy reasoning paths. To address these problems, we propose a multi-hop reasoning model with dual agents based on hierarchical reinforcement learning (HRL), which is named FULORA. FULORA tackles the above reasoning challenges by eFficient GUidance-ExpLORAtion between dual agents. The high-level agent walks on the simplified knowledge graph to provide stage-wise hints for the low-level agent walking on the original knowledge graph. In this framework, the low-level agent optimizes a value function that balances two objectives: (1) maximizing return, and (2) integrating efficient guidance from the high-level agent. Experiments conducted on three real-word knowledge graph datasets demonstrate that FULORA outperforms RL-based baselines, especially in the case of long-distance reasoning.
We present the Bio-Silicon Intelligence System (BSIS), an innovative hybrid platform that integrates biological neural networks with silicon-based computing. The BSIS, a Physics-Informed Hybrid Hierarchical Reinforcement Learning State Machine, employs carbon nanotube-coated electrodes to interface rat brains with computational systems, enabling high-fidelity neural interfacing and bidirectional communication through self-organizing systems in both biological and silicon forms. Our system leverages both analogue and digital AI theory, incorporating concepts from computational theory, chaos theory, dynamical systems theory, physics, and quantum mechanics. Additionally, the BSIS replicates the neuronal dynamics typical of intelligent brain tissue, employing nonlinear operations underlying learning and information storage. Neural signals are read through the FreeEEG32 board and BrainFlow software, then features are extracted and mapped to game actions by tracking feature changes in continuous data. Metadata is encoded into both analogue and digital brain stimulation signals at the microvolt level using our proprietary software and hardware. The system employs a dual signaling approach for training the rat brain, incorporating a reward solution and sound as well as human-inaudible distress sounds. This paper details the design, theory, functionality, and technical specifications of the BSIS, highlighting its interdisciplinary approach and advanced technological integration.
Pathology image are essential for accurately interpreting lesion cells in cytopathology screening, but acquiring high-resolution digital slides requires specialized equipment and long scanning times. Though super-resolution (SR) techniques can alleviate this problem, existing deep learning models recover pathology image in a black-box manner, which can lead to untruthful biological details and misdiagnosis. Additionally, current methods allocate the same computational resources to recover each pixel of pathology image, leading to the sub-optimal recovery issue due to the large variation of pathology image. In this paper, we propose the first hierarchical reinforcement learning framework named Spatial-Temporal hierARchical Reinforcement Learning (STAR-RL), mainly for addressing the aforementioned issues in pathology image super-resolution problem. We reformulate the SR problem as a Markov decision process of interpretable operations and adopt the hierarchical recovery mechanism in patch level, to avoid sub-optimal recovery. Specifically, the higher-level spatial manager is proposed to pick out the most corrupted patch for the lower-level patch worker. Moreover, the higher-level temporal manager is advanced to evaluate the selected patch and determine whether the optimization should be stopped earlier, thereby avoiding the over-processed problem. Under the guidance of spatial-temporal managers, the lower-level patch worker processes the selected patch with pixel-wise interpretable actions at each time step. Experimental results on medical images degraded by different kernels show the effectiveness of STAR-RL. Furthermore, STAR-RL validates the promotion in tumor diagnosis with a large margin and shows generalizability under various degradations. The source code is available at https://github.com/CUHK-AIM-Group/STAR-RL.
Hierarchical reinforcement learning (HRL) addresses complex long-horizon tasks by skillfully decomposing them into subgoals. Therefore, the effectiveness of HRL is greatly influenced by subgoal reachability. Typical HRL methods only consider subgoal reachability from the unilateral level, where a dominant level enforces compliance to the subordinate level. However, we observe that when the dominant level becomes trapped in local exploration or generates unattainable subgoals, the subordinate level is negatively affected and cannot follow the dominant level's actions. This can potentially make both levels stuck in local optima, ultimately hindering subsequent subgoal reachability. Allowing real-time bilateral information sharing and error correction would be a natural cure for this issue, which motivates us to propose a mutual response mechanism. Based on this, we propose the Bidirectional-reachable Hierarchical Policy Optimization (BrHPO)--a simple yet effective algorithm that also enjoys computation efficiency. Experiment results on a variety of long-horizon tasks showcase that BrHPO outperforms other state-of-the-art HRL baselines, coupled with a significantly higher exploration efficiency and robustness.
As an essential resource management problem in network virtualization, virtual network embedding (VNE) aims to allocate the finite resources of physical network to sequentially arriving virtual network requests (VNRs) with different resource demands. Since this is an NP-hard combinatorial optimization problem, many efforts have been made to provide viable solutions. However, most existing approaches have either ignored the admission control of VNRs, which has a potential impact on long-term performances, or not fully exploited the temporal and topological features of the physical network and VNRs. In this paper, we propose a deep Hierarchical Reinforcement Learning approach to learn a joint Admission Control and Resource Allocation policy for VNE, named HRL-ACRA. Specifically, the whole VNE process is decomposed into an upper-level policy for deciding whether to admit the arriving VNR or not and a lower-level policy for allocating resources of the physical network to meet the requirement of VNR through the HRL approach. Considering the proximal policy optimization as the basic training algorithm, we also adopt the average reward method to address the infinite horizon problem of the upper-level agent and design a customized multi-objective intrinsic reward to alleviate the sparse reward issue of the lower-level agent. Moreover, we develop a deep feature-aware graph neural network to capture the features of VNR and physical network and exploit a sequence-to-sequence model to generate embedding actions iteratively. Finally, extensive experiments are conducted in various settings, and show that HRL-ACRA outperforms state-of-the-art baselines in terms of both the acceptance ratio and long-term average revenue. Our code is available at \url{https://github.com/GeminiLight/hrl-acra}.
Knowledge-enhanced pre-trained language models (KEPLMs) leverage relation triples from knowledge graphs (KGs) and integrate these external data sources into language models via self-supervised learning. Previous works treat knowledge enhancement as two independent operations, i.e., knowledge injection and knowledge integration. In this paper, we propose to learn Knowledge-Enhanced language representations with Hierarchical Reinforcement Learning (KEHRL), which jointly addresses the problems of detecting positions for knowledge injection and integrating external knowledge into the model in order to avoid injecting inaccurate or irrelevant knowledge. Specifically, a high-level reinforcement learning (RL) agent utilizes both internal and prior knowledge to iteratively detect essential positions in texts for knowledge injection, which filters out less meaningful entities to avoid diverting the knowledge learning direction. Once the entity positions are selected, a relevant triple filtration module is triggered to perform low-level RL to dynamically refine the triples associated with polysemic entities through binary-valued actions. Experiments validate KEHRL's effectiveness in probing factual knowledge and enhancing the model's performance on various natural language understanding tasks.
In goal-conditioned hierarchical reinforcement learning (HRL), a high-level policy specifies a subgoal for the low-level policy to reach. Effective HRL hinges on a suitable subgoal represen tation function, abstracting state space into latent subgoal space and inducing varied low-level behaviors. Existing methods adopt a subgoal representation that provides a deterministic mapping from state space to latent subgoal space. Instead, this paper utilizes Gaussian Processes (GPs) for the first probabilistic subgoal representation. Our method employs a GP prior on the latent subgoal space to learn a posterior distribution over the subgoal representation functions while exploiting the long-range correlation in the state space through learnable kernels. This enables an adaptive memory that integrates long-range subgoal information from prior planning steps allowing to cope with stochastic uncertainties. Furthermore, we propose a novel learning objective to facilitate the simultaneous learning of probabilistic subgoal representations and policies within a unified framework. In experiments, our approach outperforms state-of-the-art baselines in standard benchmarks but also in environments with stochastic elements and under diverse reward conditions. Additionally, our model shows promising capabilities in transferring low-level policies across different tasks.