ReMA: Learning to Meta-think for LLMs with Multi-Agent Reinforcement Learning

Recent research on Reasoning of Large Language Models (LLMs) has sought to further enhance their performance by integrating meta-thinking -- enabling models to monitor, evaluate, and control their reasoning processes for more adaptive and effective problem-solving. However, current single-agent work lacks a specialized design for acquiring meta-thinking, resulting in low efficacy. To address this challenge, we introduce Reinforced Meta-thinking Agents (ReMA), a novel framework that leverages Multi-Agent Reinforcement Learning (MARL) to elicit meta-thinking behaviors, encouraging LLMs to think about thinking. ReMA decouples the reasoning process into two hierarchical agents: a high-level meta-thinking agent responsible for generating strategic oversight and plans, and a low-level reasoning agent for detailed executions. Through iterative reinforcement learning with aligned objectives, these agents explore and learn collaboration, leading to improved generalization and robustness. Experimental results demonstrate that ReMA outperforms single-agent RL baselines on complex reasoning tasks, including competitive-level mathematical benchmarks and LLM-as-a-Judge benchmarks. Comprehensive ablation studies further illustrate the evolving dynamics of each distinct agent, providing valuable insights into how the meta-thinking reasoning process enhances the reasoning capabilities of LLMs.

Adding Alignment Control to Language Models

Post-training alignment has increasingly become a crucial factor in enhancing the usability of language models (LMs). However, the strength of alignment varies depending on individual preferences. This paper proposes a method to incorporate alignment control into a single model, referred to as CLM. This approach adds one identity layer preceding the initial layers and performs preference learning only on this layer to map unaligned input token embeddings into the aligned space. Experimental results demonstrate that this efficient fine-tuning method performs comparable to full fine-tuning. During inference, the input embeddings are processed through the aligned and unaligned layers, which are then merged through the interpolation coefficient. By controlling this parameter, the alignment exhibits a clear interpolation and extrapolation phenomenon.

PALo: Learning Posture-Aware Locomotion for Quadruped Robots

With the rapid development of embodied intelligence, locomotion control of quadruped robots on complex terrains has become a research hotspot. Unlike traditional locomotion control approaches focusing solely on velocity tracking, we pursue to balance the agility and robustness of quadruped robots on diverse and complex terrains. To this end, we propose an end-to-end deep reinforcement learning framework for posture-aware locomotion named PALo, which manages to handle simultaneous linear and angular velocity tracking and real-time adjustments of body height, pitch, and roll angles. In PALo, the locomotion control problem is formulated as a partially observable Markov decision process, and an asymmetric actor-critic architecture is adopted to overcome the sim-to-real challenge. Further, by incorporating customized training curricula, PALo achieves agile posture-aware locomotion control in simulated environments and successfully transfers to real-world settings without fine-tuning, allowing real-time control of the quadruped robot's locomotion and body posture across challenging terrains. Through in-depth experimental analysis, we identify the key components of PALo that contribute to its performance, further validating the effectiveness of the proposed method. The results of this study provide new possibilities for the low-level locomotion control of quadruped robots in higher dimensional command spaces and lay the foundation for future research on upper-level modules for embodied intelligence.

Humanoid Whole-Body Locomotion on Narrow Terrain via Dynamic Balance and Reinforcement Learning

Humans possess delicate dynamic balance mechanisms that enable them to maintain stability across diverse terrains and under extreme conditions. However, despite significant advances recently, existing locomotion algorithms for humanoid robots are still struggle to traverse extreme environments, especially in cases that lack external perception (e.g., vision or LiDAR). This is because current methods often rely on gait-based or perception-condition rewards, lacking effective mechanisms to handle unobservable obstacles and sudden balance loss. To address this challenge, we propose a novel whole-body locomotion algorithm based on dynamic balance and Reinforcement Learning (RL) that enables humanoid robots to traverse extreme terrains, particularly narrow pathways and unexpected obstacles, using only proprioception. Specifically, we introduce a dynamic balance mechanism by leveraging an extended measure of Zero-Moment Point (ZMP)-driven rewards and task-driven rewards in a whole-body actor-critic framework, aiming to achieve coordinated actions of the upper and lower limbs for robust locomotion. Experiments conducted on a full-sized Unitree H1-2 robot verify the ability of our method to maintain balance on extremely narrow terrains and under external disturbances, demonstrating its effectiveness in enhancing the robot's adaptability to complex environments. The videos are given at https://whole-body-loco.github.io.

ThinkBench: Dynamic Out-of-Distribution Evaluation for Robust LLM Reasoning

Evaluating large language models (LLMs) poses significant challenges, particularly due to issues of data contamination and the leakage of correct answers. To address these challenges, we introduce ThinkBench, a novel evaluation framework designed to evaluate LLMs' reasoning capability robustly. ThinkBench proposes a dynamic data generation method for constructing out-of-distribution (OOD) datasets and offers an OOD dataset that contains 2,912 samples drawn from reasoning tasks. ThinkBench unifies the evaluation of reasoning models and non-reasoning models. We evaluate 16 LLMs and 4 PRMs under identical experimental conditions and show that most of the LLMs' performance are far from robust and they face a certain level of data leakage. By dynamically generating OOD datasets, ThinkBench effectively provides a reliable evaluation of LLMs and reduces the impact of data contamination.

Retrieval-Augmented Process Reward Model for Generalizable Mathematical Reasoning

While large language models (LLMs) have significantly advanced mathematical reasoning, Process Reward Models (PRMs) have been developed to evaluate the logical validity of reasoning steps. However, PRMs still struggle with out-of-distribution (OOD) challenges. This paper identifies key OOD issues, including step OOD, caused by differences in reasoning patterns across model types and sizes, and question OOD, which arises from dataset shifts between training data and real-world problems. To address these issues, we introduce Retrieval-Augmented Process Reward Model (RetrievalPRM), a novel framework designed to tackle these OOD issues. By utilizing a two-stage retrieval-enhanced mechanism, RetrievalPRM retrieves semantically similar questions and steps as a warmup, enhancing PRM's ability to evaluate target steps and improving generalization and reasoning consistency across different models and problem types. Our extensive experiments demonstrate that RetrievalPRM outperforms existing baselines across multiple real-world datasets. Our open-source contributions include a retrieval-enhanced dataset, a tuning framework for PRM training, and the RetrievalPRM model, establishing a new standard for PRM performance.

Bursting Filter Bubble: Enhancing Serendipity Recommendations with Aligned Large Language Models

Recommender systems (RSs) often suffer from the feedback loop phenomenon, e.g., RSs are trained on data biased by their recommendations. This leads to the filter bubble effect that reinforces homogeneous content and reduces user satisfaction. To this end, serendipity recommendations, which offer unexpected yet relevant items, are proposed. Recently, large language models (LLMs) have shown potential in serendipity prediction due to their extensive world knowledge and reasoning capabilities. However, they still face challenges in aligning serendipity judgments with human assessments, handling long user behavior sequences, and meeting the latency requirements of industrial RSs. To address these issues, we propose SERAL (Serendipity Recommendations with Aligned Large Language Models), a framework comprising three stages: (1) Cognition Profile Generation to compress user behavior into multi-level profiles; (2) SerenGPT Alignment to align serendipity judgments with human preferences using enriched training data; and (3) Nearline Adaptation to integrate SerenGPT into industrial RSs pipelines efficiently. Online experiments demonstrate that SERAL improves exposure ratio (PVR), clicks, and transactions of serendipitous items by 5.7%, 29.56%, and 27.6%, enhancing user experience without much impact on overall revenue. Now, it has been fully deployed in the "Guess What You Like" of the Taobao App homepage.

RHINO: Learning Real-Time Humanoid-Human-Object Interaction from Human Demonstrations

Humanoid robots have shown success in locomotion and manipulation. Despite these basic abilities, humanoids are still required to quickly understand human instructions and react based on human interaction signals to become valuable assistants in human daily life. Unfortunately, most existing works only focus on multi-stage interactions, treating each task separately, and neglecting real-time feedback. In this work, we aim to empower humanoid robots with real-time reaction abilities to achieve various tasks, allowing human to interrupt robots at any time, and making robots respond to humans immediately. To support such abilities, we propose a general humanoid-human-object interaction framework, named RHINO, i.e., Real-time Humanoid-human Interaction and Object manipulation. RHINO provides a unified view of reactive motion, instruction-based manipulation, and safety concerns, over multiple human signal modalities, such as languages, images, and motions. RHINO is a hierarchical learning framework, enabling humanoids to learn reaction skills from human-human-object demonstrations and teleoperation data. In particular, it decouples the interaction process into two levels: 1) a high-level planner inferring human intentions from real-time human behaviors; and 2) a low-level controller achieving reactive motion behaviors and object manipulation skills based on the predicted intentions. We evaluate the proposed framework on a real humanoid robot and demonstrate its effectiveness, flexibility, and safety in various scenarios.

Boost, Disentangle, and Customize: A Robust System2-to-System1 Pipeline for Code Generation

Large language models (LLMs) have demonstrated remarkable capabilities in various domains, particularly in system 1 tasks, yet the intricacies of their problem-solving mechanisms in system 2 tasks are not sufficiently explored. Recent research on System2-to-System1 methods surge, exploring the System 2 reasoning knowledge via inference-time computation and compressing the explored knowledge into System 1 process. In this paper, we focus on code generation, which is a representative System 2 task, and identify two primary challenges: (1) the complex hidden reasoning processes and (2) the heterogeneous data distributions that complicate the exploration and training of robust LLM solvers. To tackle these issues, we propose a novel BDC framework that explores insightful System 2 knowledge of LLMs using a MC-Tree-Of-Agents algorithm with mutual \textbf{B}oosting, \textbf{D}isentangles the heterogeneous training data for composable LoRA-experts, and obtain \textbf{C}ustomized problem solver for each data instance with an input-aware hypernetwork to weight over the LoRA-experts, offering effectiveness, flexibility, and robustness. This framework leverages multiple LLMs through mutual verification and boosting, integrated into a Monte-Carlo Tree Search process enhanced by reflection-based pruning and refinement. Additionally, we introduce the DisenLora algorithm, which clusters heterogeneous data to fine-tune LLMs into composable Lora experts, enabling the adaptive generation of customized problem solvers through an input-aware hypernetwork. This work lays the groundwork for advancing LLM capabilities in complex reasoning tasks, offering a novel System2-to-System1 solution.

Score-Based Diffusion Policy Compatible with Reinforcement Learning via Optimal Transport

Diffusion policies have shown promise in learning complex behaviors from demonstrations, particularly for tasks requiring precise control and long-term planning. However, they face challenges in robustness when encountering distribution shifts. This paper explores improving diffusion-based imitation learning models through online interactions with the environment. We propose OTPR (Optimal Transport-guided score-based diffusion Policy for Reinforcement learning fine-tuning), a novel method that integrates diffusion policies with RL using optimal transport theory. OTPR leverages the Q-function as a transport cost and views the policy as an optimal transport map, enabling efficient and stable fine-tuning. Moreover, we introduce masked optimal transport to guide state-action matching using expert keypoints and a compatibility-based resampling strategy to enhance training stability. Experiments on three simulation tasks demonstrate OTPR's superior performance and robustness compared to existing methods, especially in complex and sparse-reward environments. In sum, OTPR provides an effective framework for combining IL and RL, achieving versatile and reliable policy learning. The code will be released at https://github.com/Sunmmyy/OTPR.git.