Hi-Agent: Hierarchical Vision-Language Agents for Mobile Device Control

Building agents that autonomously operate mobile devices has attracted increasing attention. While Vision-Language Models (VLMs) show promise, most existing approaches rely on direct state-to-action mappings, which lack structured reasoning and planning, and thus generalize poorly to novel tasks or unseen UI layouts. We introduce Hi-Agent, a trainable hierarchical vision-language agent for mobile control, featuring a high-level reasoning model and a low-level action model that are jointly optimized. For efficient training, we reformulate multi-step decision-making as a sequence of single-step subgoals and propose a foresight advantage function, which leverages execution feedback from the low-level model to guide high-level optimization. This design alleviates the path explosion issue encountered by Group Relative Policy Optimization (GRPO) in long-horizon tasks and enables stable, critic-free joint training. Hi-Agent achieves a new State-Of-The-Art (SOTA) 87.9% task success rate on the Android-in-the-Wild (AitW) benchmark, significantly outperforming prior methods across three paradigms: prompt-based (AppAgent: 17.7%), supervised (Filtered BC: 54.5%), and reinforcement learning-based (DigiRL: 71.9%). It also demonstrates competitive zero-shot generalization on the ScreenSpot-v2 benchmark. On the more challenging AndroidWorld benchmark, Hi-Agent also scales effectively with larger backbones, showing strong adaptability in high-complexity mobile control scenarios.

Optimistic Reinforcement Learning-Based Skill Insertions for Task and Motion Planning

Task and motion planning (TAMP) for robotics manipulation necessitates long-horizon reasoning involving versatile actions and skills. While deterministic actions can be crafted by sampling or optimizing with certain constraints, planning actions with uncertainty, i.e., probabilistic actions, remains a challenge for TAMP. On the contrary, Reinforcement Learning (RL) excels in acquiring versatile, yet short-horizon, manipulation skills that are robust with uncertainties. In this letter, we design a method that integrates RL skills into TAMP pipelines. Besides the policy, a RL skill is defined with data-driven logical components that enable the skill to be deployed by symbolic planning. A plan refinement sub-routine is designed to further tackle the inevitable effect uncertainties. In the experiments, we compare our method with baseline hierarchical planning from both TAMP and RL fields and illustrate the strength of the method. The results show that by embedding RL skills, we extend the capability of TAMP to domains with probabilistic skills, and improve the planning efficiency compared to the previous methods.

An LLM-Powered Cooperative Framework for Large-Scale Multi-Vehicle Navigation

The rise of Internet of Vehicles (IoV) technologies is transforming traffic management from isolated control to a collective, multi-vehicle process. At the heart of this shift is multi-vehicle dynamic navigation, which requires simultaneously routing large fleets under evolving traffic conditions. Existing path search algorithms and reinforcement learning methods struggle to scale to city-wide networks, often failing to capture the nonlinear, stochastic, and coupled dynamics of urban traffic. To address these challenges, we propose CityNav, a hierarchical, LLM-powered framework for large-scale multi-vehicle navigation. CityNav integrates a global traffic allocation agent, which coordinates strategic traffic flow distribution across regions, with local navigation agents that generate locally adaptive routes aligned with global directives. To enable effective cooperation, we introduce a cooperative reasoning optimization mechanism, in which agents are jointly trained with a dual-reward structure: individual rewards promote per-vehicle efficiency, while shared rewards encourage network-wide coordination and congestion reduction. Extensive experiments on four real-world road networks of varying scales (up to 1.6 million roads and 430,000 intersections) and traffic datasets demonstrate that CityNav consistently outperforms nine classical path search and RL-based baselines in city-scale travel efficiency and congestion mitigation. Our results highlight the potential of LLMs to enable scalable, adaptive, and cooperative city-wide traffic navigation, providing a foundation for intelligent, large-scale vehicle routing in complex urban environments. Our project is available at https://github.com/usail-hkust/CityNav.

Reasoning for Hierarchical Text Classification: The Case of Patents

Hierarchical text classification (HTC) assigns documents to multiple levels of a pre-defined taxonomy. Automated patent subject classification represents one of the hardest HTC scenarios because of domain knowledge difficulty and a huge number of labels. Prior approaches only output a flat label set, which offers little insight into the reason behind predictions. Therefore, we propose Reasoning for Hierarchical Classification (RHC), a novel framework that reformulates HTC as a step-by-step reasoning task to sequentially deduce hierarchical labels. RHC trains large language models (LLMs) in two stages: a cold-start stage that aligns outputs with chain-of-thought (CoT) reasoning format and a reinforcement learning (RL) stage to enhance multi-step reasoning ability. RHC demonstrates four advantages in our experiments. (1) Effectiveness: RHC surpasses previous baselines and outperforms the supervised fine-tuning counterparts by approximately 3% in accuracy and macro F1. (2) Explainability: RHC produces natural-language justifications before prediction to facilitate human inspection. (3) Scalability: RHC scales favorably with model size with larger gains compared to standard fine-tuning. (4) Applicability: Beyond patents, we further demonstrate that RHC achieves state-of-the-art performance on other widely used HTC benchmarks, which highlights its broad applicability.

PyramidStyler: Transformer-Based Neural Style Transfer with Pyramidal Positional Encoding and Reinforcement Learning

Neural Style Transfer (NST) has evolved from Gatys et al.'s (2015) CNN-based algorithm, enabling AI-driven artistic image synthesis. However, existing CNN and transformer-based models struggle to scale efficiently to complex styles and high-resolution inputs. We introduce PyramidStyler, a transformer framework with Pyramidal Positional Encoding (PPE): a hierarchical, multi-scale encoding that captures both local details and global context while reducing computational load. We further incorporate reinforcement learning to dynamically optimize stylization, accelerating convergence. Trained on Microsoft COCO and WikiArt, PyramidStyler reduces content loss by 62.6% (to 2.07) and style loss by 57.4% (to 0.86) after 4000 epochs--achieving 1.39 s inference--and yields further improvements (content 2.03; style 0.75) with minimal speed penalty (1.40 s) when using RL. These results demonstrate real-time, high-quality artistic rendering, with broad applications in media and design.

Step-Aware Policy Optimization for Reasoning in Diffusion Large Language Models

Diffusion language models (dLLMs) offer a promising, non-autoregressive paradigm for text generation, yet training them for complex reasoning remains a key challenge. Current reinforcement learning approaches often rely on sparse, outcome-based rewards, which can reinforce flawed reasoning paths that lead to coincidentally correct answers. We argue that this stems from a fundamental mismatch with the natural structure of reasoning. We first propose a theoretical framework that formalizes complex problem solving as a hierarchical selection process, where an intractable global constraint is decomposed into a series of simpler, localized logical steps. This framework provides a principled foundation for algorithm design, including theoretical insights into the identifiability of this latent reasoning structure. Motivated by this theory, we identify unstructured refinement -- a failure mode where a model's iterative steps do not contribute meaningfully to the solution -- as a core deficiency in existing methods. We then introduce Step-Aware Policy Optimization (SAPO), a novel RL algorithm that aligns the dLLM's denoising process with the latent reasoning hierarchy. By using a process-based reward function that encourages incremental progress, SAPO guides the model to learn structured, coherent reasoning paths. Our empirical results show that this principled approach significantly improves performance on challenging reasoning benchmarks and enhances the interpretability of the generation process.

Retargeting Matters: General Motion Retargeting for Humanoid Motion Tracking

Humanoid motion tracking policies are central to building teleoperation pipelines and hierarchical controllers, yet they face a fundamental challenge: the embodiment gap between humans and humanoid robots. Current approaches address this gap by retargeting human motion data to humanoid embodiments and then training reinforcement learning (RL) policies to imitate these reference trajectories. However, artifacts introduced during retargeting, such as foot sliding, self-penetration, and physically infeasible motion are often left in the reference trajectories for the RL policy to correct. While prior work has demonstrated motion tracking abilities, they often require extensive reward engineering and domain randomization to succeed. In this paper, we systematically evaluate how retargeting quality affects policy performance when excessive reward tuning is suppressed. To address issues that we identify with existing retargeting methods, we propose a new retargeting method, General Motion Retargeting (GMR). We evaluate GMR alongside two open-source retargeters, PHC and ProtoMotions, as well as with a high-quality closed-source dataset from Unitree. Using BeyondMimic for policy training, we isolate retargeting effects without reward tuning. Our experiments on a diverse subset of the LAFAN1 dataset reveal that while most motions can be tracked, artifacts in retargeted data significantly reduce policy robustness, particularly for dynamic or long sequences. GMR consistently outperforms existing open-source methods in both tracking performance and faithfulness to the source motion, achieving perceptual fidelity and policy success rates close to the closed-source baseline. Website: https://jaraujo98.github.io/retargeting_matters. Code: https://github.com/YanjieZe/GMR.

ReLAM: Learning Anticipation Model for Rewarding Visual Robotic Manipulation

Reward design remains a critical bottleneck in visual reinforcement learning (RL) for robotic manipulation. In simulated environments, rewards are conventionally designed based on the distance to a target position. However, such precise positional information is often unavailable in real-world visual settings due to sensory and perceptual limitations. In this study, we propose a method that implicitly infers spatial distances through keypoints extracted from images. Building on this, we introduce Reward Learning with Anticipation Model (ReLAM), a novel framework that automatically generates dense, structured rewards from action-free video demonstrations. ReLAM first learns an anticipation model that serves as a planner and proposes intermediate keypoint-based subgoals on the optimal path to the final goal, creating a structured learning curriculum directly aligned with the task's geometric objectives. Based on the anticipated subgoals, a continuous reward signal is provided to train a low-level, goal-conditioned policy under the hierarchical reinforcement learning (HRL) framework with provable sub-optimality bound. Extensive experiments on complex, long-horizon manipulation tasks show that ReLAM significantly accelerates learning and achieves superior performance compared to state-of-the-art methods.

Structural Information-based Hierarchical Diffusion for Offline Reinforcement Learning

Diffusion-based generative methods have shown promising potential for modeling trajectories from offline reinforcement learning (RL) datasets, and hierarchical diffusion has been introduced to mitigate variance accumulation and computational challenges in long-horizon planning tasks. However, existing approaches typically assume a fixed two-layer diffusion hierarchy with a single predefined temporal scale, which limits adaptability to diverse downstream tasks and reduces flexibility in decision making. In this work, we propose SIHD, a novel Structural Information-based Hierarchical Diffusion framework for effective and stable offline policy learning in long-horizon environments with sparse rewards. Specifically, we analyze structural information embedded in offline trajectories to construct the diffusion hierarchy adaptively, enabling flexible trajectory modeling across multiple temporal scales. Rather than relying on reward predictions from localized sub-trajectories, we quantify the structural information gain of each state community and use it as a conditioning signal within the corresponding diffusion layer. To reduce overreliance on offline datasets, we introduce a structural entropy regularizer that encourages exploration of underrepresented states while avoiding extrapolation errors from distributional shifts. Extensive evaluations on challenging offline RL tasks show that SIHD significantly outperforms state-of-the-art baselines in decision-making performance and demonstrates superior generalization across diverse scenarios.

Hierarchical Reinforcement Learning with Low-Level MPC for Multi-Agent Control

Achieving safe and coordinated behavior in dynamic, constraint-rich environments remains a major challenge for learning-based control. Pure end-to-end learning often suffers from poor sample efficiency and limited reliability, while model-based methods depend on predefined references and struggle to generalize. We propose a hierarchical framework that combines tactical decision-making via reinforcement learning (RL) with low-level execution through Model Predictive Control (MPC). For the case of multi-agent systems this means that high-level policies select abstract targets from structured regions of interest (ROIs), while MPC ensures dynamically feasible and safe motion. Tested on a predator-prey benchmark, our approach outperforms end-to-end and shielding-based RL baselines in terms of reward, safety, and consistency, underscoring the benefits of combining structured learning with model-based control.