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 (Grounding Language Models as EffIcient Decision-Making Agents via Offline HiErarchical Reinforcement 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-context reinforcement learning (ICRL) has emerged as a promising paradigm for adapting RL agents to downstream tasks through prompt conditioning. However, two notable challenges remain in fully harnessing in-context learning within RL domains: the intrinsic multi-modality of the state-action-reward data and the diverse, heterogeneous nature of decision tasks. To tackle these challenges, we propose **T2MIR** (**T**oken- and **T**ask-wise **M**oE for **I**n-context **R**L), an innovative framework that introduces architectural advances of mixture-of-experts (MoE) into transformer-based decision models. T2MIR substitutes the feedforward layer with two parallel layers: a token-wise MoE that captures distinct semantics of input tokens across multiple modalities, and a task-wise MoE that routes diverse tasks to specialized experts for managing a broad task distribution with alleviated gradient conflicts. To enhance task-wise routing, we introduce a contrastive learning method that maximizes the mutual information between the task and its router representation, enabling more precise capture of task-relevant information. The outputs of two MoE components are concatenated and fed into the next layer. Comprehensive experiments show that T2MIR significantly facilitates in-context learning capacity and outperforms various types of baselines. We bring the potential and promise of MoE to ICRL, offering a simple and scalable architectural enhancement to advance ICRL one step closer toward achievements in language and vision communities. Our code is available at [https://github.com/NJU-RL/T2MIR](https://github.com/NJU-RL/T2MIR).

Recent advances in large reasoning models (LRMs) demonstrate that sophisticated behaviors such as multi-step reasoning and self-reflection can emerge via reinforcement learning with verifiable rewards~(RLVR). However, existing RLVR approaches are inherently ``on-policy'', limiting learning to a model's own outputs and failing to acquire reasoning abilities beyond its initial capabilities. To address this issue, we introduce LUFFY (Learning to reason Under oFF-policY guidance), a framework that augments RLVR with off-policy reasoning traces. LUFFY dynamically balances imitation and exploration by combining off-policy demonstrations with on-policy rollouts during training. Specifically, LUFFY combines the Mixed-Policy GRPO framework, which has a theoretically guaranteed convergence rate, alongside policy shaping via regularized importance sampling to avoid superficial and rigid imitation during mixed-policy training. Compared with previous RLVR methods, LUFFY achieves an over +6.4 average gain across six math benchmarks and an advantage of over +6.2 points in out-of-distribution tasks. Most significantly, we show that LUFFY successfully trains weak models in scenarios where on-policy RLVR completely fails. These results provide compelling evidence that LUFFY transcends the fundamental limitations of on-policy RLVR and demonstrates the great potential of utilizing off-policy guidance in RLVR.

Offline meta-RL usually tackles generalization by inferring task beliefs from high-quality samples or warmup explorations. The restricted form limits their generality and usability since these supervision signals are expensive and even infeasible to acquire in advance for unseen tasks. Learning directly from the raw text about decision tasks is a promising alternative to leverage a much broader source of supervision. In the paper, we propose **T**ext-to-**D**ecision **A**gent (**T2DA**), a simple and scalable framework that supervises offline meta-RL with natural language. We first introduce a generalized world model to encode multi-task decision data into a dynamics-aware embedding space. Then, inspired by CLIP, we predict which textual description goes with which decision embedding, effectively bridging their semantic gap via contrastive language-decision pre-training and aligning the text embeddings to comprehend the environment dynamics. After training the text-conditioned generalist policy, the agent can directly realize zero-shot text-to-decision generation in response to language instructions. Comprehensive experiments on MuJoCo and Meta-World benchmarks show that T2DA facilitates high-capacity zero-shot generalization and outperforms various types of baselines. Our code is available at [https://github.com/NJU-RL/T2DA](https://github.com/NJU-RL/T2DA).

Inspired by diffusion models that revolutionized image generation through language conditioning, offline reinforcement learning (RL) has been reformulated as a sequence modeling problem. Similar to how language descriptions guide image generation, RL sequence models require task-specific conditioning information to generalize to new tasks. We propose Prompt Decision Diffuser (PDD), addressing the generalization challenge in offline meta reinforcement learning (OMRL) by treating it as a sequence modeling problem, with demonstration-based prompting for few-shot adaptation. These prompts, encoded from few-shot demonstrations, effectively capture task-specific information to guide cross-task policy generation. Experimental results on Mujoco and Point-Robot benchmarks demonstrate PDD's superior few-shot generalization capabilities compared to baseline approaches.

Real-world multi-agent tasks usually involve dynamic team composition with the emergence of roles, which should also be a key to efficient cooperation in multi-agent reinforcement learning (MARL). Drawing inspiration from the correlation between roles and agent's behavior patterns, we propose a novel framework of **A**ttention-guided **CO**ntrastive **R**ole representation learning for **M**ARL (**ACORM**) to promote behavior heterogeneity, knowledge transfer, and skillful coordination across agents. First, we introduce mutual information maximization to formalize role representation learning, derive a contrastive learning objective, and concisely approximate the distribution of negative pairs. Second, we leverage an attention mechanism to prompt the global state to attend to learned role representations in value decomposition, implicitly guiding agent coordination in a skillful role space to yield more expressive credit assignment. Experiments on challenging StarCraft II micromanagement and Google research football tasks demonstrate the state-of-the-art performance of our method and its advantages over existing approaches. Our code is available at [https://github.com/NJU-RL/ACORM](https://github.com/NJU-RL/ACORM).

Road detection in remote sensing images holds significant application value in urban planning and autonomous driving. However, due to the complex background and the strip-like distribution of roads, road detection remains a challenging task. In this letter, we propose an innovative method for remote sensing road detection called Direction-Aware and Foreground-Guided (DAFG). Firstly, direction-aware trans bridge (DATB) is formulated, which uses a dual-branch structure encompassing vertical and horizontal dimensions to capture the global-local context, effectively mitigating interference from non-road area features. Next, foreground guidance module (FGM) is constructed, with the help of prediction maps derived from high-level semantic features, enhancing the encoding of low-level road features and suppressing background and noise. Finally, a dense aggregation module (DAM) is introduced to enhance feature representation and augment the capacity of classic decoding blocks (several cascaded convolutions). Experiments show that DAFG achieves higher intersection over union (IoU) than other methods on the Munich Road dataset (89.22%) and the WHU Road dataset (70.78%).

Road detection based on remote sensing images is of great significance to intelligent traffic management. The performances of the mainstream road detection methods are mainly determined by their extracted features, whose richness and robustness can be enhanced by fusing features of different types and cross-layer connections. However, the features in the existing mainstream model frameworks are often similar in the same layer by the single-task training, and the traditional cross-layer fusion ways are too simple to obtain an efficient effect, so more complex fusion ways besides concatenation and addition deserve to be explored. Aiming at the above defects, we propose a dual-task network (DTnet) for road detection and cross-layer graph fusion module (CGM): the DTnet consists of two parallel branches for road area and edge detection, respectively, while enhancing the feature diversity by fusing features between two branches through our designed feature bridge modules (FBM). The CGM improves the cross-layer fusion effect by a complex feature stream graph, and four graph patterns are evaluated. Experimental results on three public datasets demonstrate that our method effectively improves the final detection result. The used code is available at https://github.com/huzican695/DTnet.

Vehicle detection based on remote sensing images is of great significance to intelligent traffic management and smart city planning. The performances of the mainstream vehicle detection methods such as Fast-RCNN, SSD, FCRN, etc. are mainly affected by their extracted features, whose richness and robustness can be enhanced by introducing pre-training models from other datasets. However, the weights of the pre-training models used in traditional methods like FCRN, only shallow-tuned from the weights of another database training, without deep optimization from random initialization. This letter proposes a new pre-training and random-initialized fusion network (PRFN): the random initialization model and the pre-training model are connected in parallel, and an encoder is used to fuse their features, thereby improving the richness of features and the robustness of the model. We evaluate four different combination patterns of the pre-training and random initialized models, and the effect of the fusion feature encoder. Experiments show the parallel concatenating pattern with the fusion feature encoder achieves the best performance than other patterns. We test on three datasets, our PRFN exceeds the traditional methods, and, respectively, 1.23%, 1.04%, 0.94% higher than the baseline. All codes are available fromhttps://github.com/LHKRobert/PRFN.