Detecting RLVR Training Data via Structural Convergence of Reasoning

Reinforcement learning with verifiable rewards (RLVR) is central to training modern reasoning models, but the undisclosed training data raises concerns about benchmark contamination. Unlike pretraining methods, which optimize models using token-level probabilities, RLVR fine-tunes models based on reward feedback from self-generated reasoning trajectories, making conventional likelihood-based detection methods less effective. We show that RLVR induces a distinctive behavioral signature: prompts encountered during RLVR training result in more rigid and similar generations, while unseen prompts retain greater diversity. We introduce Min-$k$NN Distance, a simple black-box detector that quantifies this collapse by sampling multiple completions for a given prompt and computing the average of the $k$ smallest nearest-neighbor edit distances. Min-$k$NN Distance requires no access to the reference model or token probabilities. Experiments across multiple RLVR-trained reasoning models show that Min-$k$NN Distance reliably distinguishes RL-seen examples from unseen ones and outperforms existing membership inference and RL contamination detection baselines.

SWE-AGI: Benchmarking Specification-Driven Software Construction with MoonBit in the Era of Autonomous Agents

Although large language models (LLMs) have demonstrated impressive coding capabilities, their ability to autonomously build production-scale software from explicit specifications remains an open question. We introduce SWE-AGI, an open-source benchmark for evaluating end-to-end, specification-driven construction of software systems written in MoonBit. SWE-AGI tasks require LLM-based agents to implement parsers, interpreters, binary decoders, and SAT solvers strictly from authoritative standards and RFCs under a fixed API scaffold. Each task involves implementing 1,000-10,000 lines of core logic, corresponding to weeks or months of engineering effort for an experienced human developer. By leveraging the nascent MoonBit ecosystem, SWE-AGI minimizes data leakage, forcing agents to rely on long-horizon architectural reasoning rather than code retrieval. Across frontier models, gpt-5.3-codex achieves the best overall performance (solving 19/22 tasks, 86.4%), outperforming claude-opus-4.6 (15/22, 68.2%), and kimi-2.5 exhibits the strongest performance among open-source models. Performance degrades sharply with increasing task difficulty, particularly on hard, specification-intensive systems. Behavioral analysis further reveals that as codebases scale, code reading, rather than writing, becomes the dominant bottleneck in AI-assisted development. Overall, while specification-driven autonomous software engineering is increasingly viable, substantial challenges remain before it can reliably support production-scale development.

The Llama 4 Herd: Architecture, Training, Evaluation, and Deployment Notes

This document consolidates publicly reported technical details about Metas Llama 4 model family. It summarizes (i) released variants (Scout and Maverick) and the broader herd context including the previewed Behemoth teacher model, (ii) architectural characteristics beyond a high-level MoE description covering routed/shared-expert structure, early-fusion multimodality, and long-context design elements reported for Scout (iRoPE and length generalization strategies), (iii) training disclosures spanning pre-training, mid-training for long-context extension, and post-training methodology (lightweight SFT, online RL, and lightweight DPO) as described in release materials, (iv) developer-reported benchmark results for both base and instruction-tuned checkpoints, and (v) practical deployment constraints observed across major serving environments, including provider-specific context limits and quantization packaging. The manuscript also summarizes licensing obligations relevant to redistribution and derivative naming, and reviews publicly described safeguards and evaluation practices. The goal is to provide a compact technical reference for researchers and practitioners who need precise, source-backed facts about Llama 4.

Spatial4D-Bench: A Versatile 4D Spatial Intelligence Benchmark

4D spatial intelligence involves perceiving and processing how objects move or change over time. Humans naturally possess 4D spatial intelligence, supporting a broad spectrum of spatial reasoning abilities. To what extent can Multimodal Large Language Models (MLLMs) achieve human-level 4D spatial intelligence? In this work, we present Spatial4D-Bench, a versatile 4D spatial intelligence benchmark designed to comprehensively assess the 4D spatial reasoning abilities of MLLMs. Unlike existing spatial intelligence benchmarks that are often small-scale or limited in diversity, Spatial4D-Bench provides a large-scale, multi-task evaluation benchmark consisting of ~40,000 question-answer pairs covering 18 well-defined tasks. We systematically organize these tasks into six cognitive categories: object understanding, scene understanding, spatial relationship understanding, spatiotemporal relationship understanding, spatial reasoning and spatiotemporal reasoning. Spatial4D-Bench thereby offers a structured and comprehensive benchmark for evaluating the spatial cognition abilities of MLLMs, covering a broad spectrum of tasks that parallel the versatility of human spatial intelligence. We benchmark various state-of-the-art open-source and proprietary MLLMs on Spatial4D-Bench and reveal their substantial limitations in a wide variety of 4D spatial reasoning aspects, such as route plan, action recognition, and physical plausibility reasoning. We hope that the findings provided in this work offer valuable insights to the community and that our benchmark can facilitate the development of more capable MLLMs toward human-level 4D spatial intelligence. More resources can be found on our project page.

Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving

End-to-end autonomous driving systems, predominantly trained through imitation learning, have demonstrated considerable effectiveness in leveraging large-scale expert driving data. Despite their success in open-loop evaluations, these systems often exhibit significant performance degradation in closed-loop scenarios due to causal confusion. This confusion is fundamentally exacerbated by the overreliance of the imitation learning paradigm on expert trajectories, which often contain unattributable noise and interfere with the modeling of causal relationships between environmental contexts and appropriate driving actions. To address this fundamental limitation, we propose Perception-Guided Self-Supervision (PGS) - a simple yet effective training paradigm that leverages perception outputs as the primary supervisory signals, explicitly modeling causal relationships in decision-making. The proposed framework aligns both the inputs and outputs of the decision-making module with perception results, such as lane centerlines and the predicted motions of surrounding agents, by introducing positive and negative self-supervision for the ego trajectory. This alignment is specifically designed to mitigate causal confusion arising from the inherent noise in expert trajectories. Equipped with perception-driven supervision, our method, built on a standard end-to-end architecture, achieves a Driving Score of 78.08 and a mean success rate of 48.64% on the challenging closed-loop Bench2Drive benchmark, significantly outperforming existing state-of-the-art methods, including those employing more complex network architectures and inference pipelines. These results underscore the effectiveness and robustness of the proposed PGS framework and point to a promising direction for addressing causal confusion and enhancing real-world generalization in autonomous driving.

Whole-Body Control Framework for Humanoid Robots with Heavy Limbs: A Model-Based Approach

Humanoid robots often face significant balance issues due to the motion of their heavy limbs. These challenges are particularly pronounced when attempting dynamic motion or operating in environments with irregular terrain. To address this challenge, this manuscript proposes a whole-body control framework for humanoid robots with heavy limbs, using a model-based approach that combines a kino-dynamics planner and a hierarchical optimization problem. The kino-dynamics planner is designed as a model predictive control (MPC) scheme to account for the impact of heavy limbs on mass and inertia distribution. By simplifying the robot's system dynamics and constraints, the planner enables real-time planning of motion and contact forces. The hierarchical optimization problem is formulated using Hierarchical Quadratic Programming (HQP) to minimize limb control errors and ensure compliance with the policy generated by the kino-dynamics planner. Experimental validation of the proposed framework demonstrates its effectiveness. The humanoid robot with heavy limbs controlled by the proposed framework can achieve dynamic walking speeds of up to 1.2~m/s, respond to external disturbances of up to 60~N, and maintain balance on challenging terrains such as uneven surfaces, and outdoor environments.

Direct Value Optimization: Improving Chain-of-Thought Reasoning in LLMs with Refined Values

We introduce Direct Value Optimization (DVO), an innovative reinforcement learning framework for enhancing large language models in complex reasoning tasks. Unlike traditional methods relying on preference labels, DVO utilizes value signals at individual reasoning steps, optimizing models via a mean squared error loss. The key benefit of DVO lies in its fine-grained supervision, circumventing the need for labor-intensive human annotations. Target values within the DVO are estimated using either Monte Carlo Tree Search or an outcome value model. Our empirical analysis on both mathematical and commonsense reasoning tasks shows that DVO consistently outperforms existing offline preference optimization techniques, even with fewer training steps. These findings underscore the importance of value signals in advancing reasoning capabilities and highlight DVO as a superior methodology under scenarios lacking explicit human preference information.

Learning to Hop for a Single-Legged Robot with Parallel Mechanism

This work presents the application of reinforcement learning to improve the performance of a highly dynamic hopping system with a parallel mechanism. Unlike serial mechanisms, parallel mechanisms can not be accurately simulated due to the complexity of their kinematic constraints and closed-loop structures. Besides, learning to hop suffers from prolonged aerial phase and the sparse nature of the rewards. To address them, we propose a learning framework to encode long-history feedback to account for the under-actuation brought by the prolonged aerial phase. In the proposed framework, we also introduce a simplified serial configuration for the parallel design to avoid directly simulating parallel structure during the training. A torque-level conversion is designed to deal with the parallel-serial conversion to handle the sim-to-real issue. Simulation and hardware experiments have been conducted to validate this framework.

Geo-ConvGRU: Geographically Masked Convolutional Gated Recurrent Unit for Bird-Eye View Segmentation

Convolutional Neural Networks (CNNs) have significantly impacted various computer vision tasks, however, they inherently struggle to model long-range dependencies explicitly due to the localized nature of convolution operations. Although Transformers have addressed limitations in long-range dependencies for the spatial dimension, the temporal dimension remains underexplored. In this paper, we first highlight that 3D CNNs exhibit limitations in capturing long-range temporal dependencies. Though Transformers mitigate spatial dimension issues, they result in a considerable increase in parameter and processing speed reduction. To overcome these challenges, we introduce a simple yet effective module, Geographically Masked Convolutional Gated Recurrent Unit (Geo-ConvGRU), tailored for Bird's-Eye View segmentation. Specifically, we substitute the 3D CNN layers with ConvGRU in the temporal module to bolster the capacity of networks for handling temporal dependencies. Additionally, we integrate a geographical mask into the Convolutional Gated Recurrent Unit to suppress noise introduced by the temporal module. Comprehensive experiments conducted on the NuScenes dataset substantiate the merits of the proposed Geo-ConvGRU, revealing that our approach attains state-of-the-art performance in Bird's-Eye View segmentation.

Direct Preference Optimization Using Sparse Feature-Level Constraints

The alignment of large language models (LLMs) with human preferences remains a key challenge. While post-training techniques like Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) have achieved notable success, they often introduce computational inefficiencies and training instability. In this paper, we propose Feature-level constrained Preference Optimization (FPO), a novel method designed to simplify the alignment process while ensuring stability. FPO leverages pre-trained Sparse Autoencoders (SAEs) and introduces feature-level constraints, allowing for efficient, sparsity-enforced alignment. Our approach enjoys efficiency by using sparse features activated in a well-trained sparse autoencoder and the quality of sequential KL divergence by using the feature-level offline reference. Experimental results on benchmark datasets demonstrate that FPO achieves a 5.08% absolute improvement in win rate with much lower computational cost compared to state-of-the-art baselines, making it a promising solution for efficient and controllable LLM alignments.