InnovatorBench: Evaluating Agents' Ability to Conduct Innovative LLM Research

AI agents could accelerate scientific discovery by automating hypothesis formation, experiment design, coding, execution, and analysis, yet existing benchmarks probe narrow skills in simplified settings. To address this gap, we introduce InnovatorBench, a benchmark-platform pair for realistic, end-to-end assessment of agents performing Large Language Model (LLM) research. It comprises 20 tasks spanning Data Construction, Filtering, Augmentation, Loss Design, Reward Design, and Scaffold Construction, which require runnable artifacts and assessment of correctness, performance, output quality, and uncertainty. To support agent operation, we develop ResearchGym, a research environment offering rich action spaces, distributed and long-horizon execution, asynchronous monitoring, and snapshot saving. We also implement a lightweight ReAct agent that couples explicit reasoning with executable planning using frontier models such as Claude-4, GPT-5, GLM-4.5, and Kimi-K2. Our experiments demonstrate that while frontier models show promise in code-driven research tasks, they struggle with fragile algorithm-related tasks and long-horizon decision making, such as impatience, poor resource management, and overreliance on template-based reasoning. Furthermore, agents require over 11 hours to achieve their best performance on InnovatorBench, underscoring the benchmark's difficulty and showing the potential of InnovatorBench to be the next generation of code-based research benchmark.

Interaction as Intelligence Part II: Asynchronous Human-Agent Rollout for Long-Horizon Task Training

Large Language Model (LLM) agents have recently shown strong potential in domains such as automated coding, deep research, and graphical user interface manipulation. However, training them to succeed on long-horizon, domain-specialized tasks remains challenging. Current methods primarily fall into two categories. The first relies on dense human annotations through behavior cloning, which is prohibitively expensive for long-horizon tasks that can take days or months. The second depends on outcome-driven sampling, which often collapses due to the rarity of valid positive trajectories on domain-specialized tasks. We introduce Apollo, a sampling framework that integrates asynchronous human guidance with action-level data filtering. Instead of requiring annotators to shadow every step, Apollo allows them to intervene only when the agent drifts from a promising trajectory, by providing prior knowledge, strategic advice, etc. This lightweight design makes it possible to sustain interactions for over 30 hours and produces valuable trajectories at a lower cost. Apollo then applies supervision control to filter out sub-optimal actions and prevent error propagation. Together, these components enable reliable and effective data collection in long-horizon environments. To demonstrate the effectiveness of Apollo, we evaluate it using InnovatorBench. Our experiments show that when applied to train the GLM-4.5 model on InnovatorBench, Apollo achieves more than a 50% improvement over the untrained baseline and a 28% improvement over a variant trained without human interaction. These results highlight the critical role of human-in-the-loop sampling and the robustness of Apollo's design in handling long-horizon, domain-specialized tasks.

DatasetResearch: Benchmarking Agent Systems for Demand-Driven Dataset Discovery

The rapid advancement of large language models has fundamentally shifted the bottleneck in AI development from computational power to data availability-with countless valuable datasets remaining hidden across specialized repositories, research appendices, and domain platforms. As reasoning capabilities and deep research methodologies continue to evolve, a critical question emerges: can AI agents transcend conventional search to systematically discover any dataset that meets specific user requirements, enabling truly autonomous demand-driven data curation? We introduce DatasetResearch, the first comprehensive benchmark evaluating AI agents' ability to discover and synthesize datasets from 208 real-world demands across knowledge-intensive and reasoning-intensive tasks. Our tri-dimensional evaluation framework reveals a stark reality: even advanced deep research systems achieve only 22% score on our challenging DatasetResearch-pro subset, exposing the vast gap between current capabilities and perfect dataset discovery. Our analysis uncovers a fundamental dichotomy-search agents excel at knowledge tasks through retrieval breadth, while synthesis agents dominate reasoning challenges via structured generation-yet both catastrophically fail on "corner cases" outside existing distributions. These findings establish the first rigorous baseline for dataset discovery agents and illuminate the path toward AI systems capable of finding any dataset in the digital universe. Our benchmark and comprehensive analysis provide the foundation for the next generation of self-improving AI systems and are publicly available at https://github.com/GAIR-NLP/DatasetResearch.

RoboTransfer: Geometry-Consistent Video Diffusion for Robotic Visual Policy Transfer

Imitation Learning has become a fundamental approach in robotic manipulation. However, collecting large-scale real-world robot demonstrations is prohibitively expensive. Simulators offer a cost-effective alternative, but the sim-to-real gap make it extremely challenging to scale. Therefore, we introduce RoboTransfer, a diffusion-based video generation framework for robotic data synthesis. Unlike previous methods, RoboTransfer integrates multi-view geometry with explicit control over scene components, such as background and object attributes. By incorporating cross-view feature interactions and global depth/normal conditions, RoboTransfer ensures geometry consistency across views. This framework allows fine-grained control, including background edits and object swaps. Experiments demonstrate that RoboTransfer is capable of generating multi-view videos with enhanced geometric consistency and visual fidelity. In addition, policies trained on the data generated by RoboTransfer achieve a 33.3% relative improvement in the success rate in the DIFF-OBJ setting and a substantial 251% relative improvement in the more challenging DIFF-ALL scenario. Explore more demos on our project page: https://horizonrobotics.github.io/robot_lab/robotransfer

SEM: Enhancing Spatial Understanding for Robust Robot Manipulation

A key challenge in robot manipulation lies in developing policy models with strong spatial understanding, the ability to reason about 3D geometry, object relations, and robot embodiment. Existing methods often fall short: 3D point cloud models lack semantic abstraction, while 2D image encoders struggle with spatial reasoning. To address this, we propose SEM (Spatial Enhanced Manipulation model), a novel diffusion-based policy framework that explicitly enhances spatial understanding from two complementary perspectives. A spatial enhancer augments visual representations with 3D geometric context, while a robot state encoder captures embodiment-aware structure through graphbased modeling of joint dependencies. By integrating these modules, SEM significantly improves spatial understanding, leading to robust and generalizable manipulation across diverse tasks that outperform existing baselines.

Deep learning automates Cobb angle measurement compared with multi-expert observers

Scoliosis, a prevalent condition characterized by abnormal spinal curvature leading to deformity, requires precise assessment methods for effective diagnosis and management. The Cobb angle is a widely used scoliosis quantification method that measures the degree of curvature between the tilted vertebrae. Yet, manual measuring of Cobb angles is time-consuming and labor-intensive, fraught with significant interobserver and intraobserver variability. To address these challenges and the lack of interpretability found in certain existing automated methods, we have created fully automated software that not only precisely measures the Cobb angle but also provides clear visualizations of these measurements. This software integrates deep neural network-based spine region detection and segmentation, spine centerline identification, pinpointing the most significantly tilted vertebrae, and direct visualization of Cobb angles on the original images. Upon comparison with the assessments of 7 expert readers, our algorithm exhibited a mean deviation in Cobb angle measurements of 4.17 degrees, notably surpassing the manual approach's average intra-reader discrepancy of 5.16 degrees. The algorithm also achieved intra-class correlation coefficients (ICC) exceeding 0.96 and Pearson correlation coefficients above 0.944, reflecting robust agreement with expert assessments and superior measurement reliability. Through the comprehensive reader study and statistical analysis, we believe this algorithm not only ensures a higher consensus with expert readers but also enhances interpretability and reproducibility during assessments. It holds significant promise for clinical application, potentially aiding physicians in more accurate scoliosis assessment and diagnosis, thereby improving patient care.

Causal Mediation Analysis with Multi-dimensional and Indirectly Observed Mediators

Causal mediation analysis (CMA) is a powerful method to dissect the total effect of a treatment into direct and mediated effects within the potential outcome framework. This is important in many scientific applications to identify the underlying mechanisms of a treatment effect. However, in many scientific applications the mediator is unobserved, but there may exist related measurements. For example, we may want to identify how changes in brain activity or structure mediate an antidepressant's effect on behavior, but we may only have access to electrophysiological or imaging brain measurements. To date, most CMA methods assume that the mediator is one-dimensional and observable, which oversimplifies such real-world scenarios. To overcome this limitation, we introduce a CMA framework that can handle complex and indirectly observed mediators based on the identifiable variational autoencoder (iVAE) architecture. We prove that the true joint distribution over observed and latent variables is identifiable with the proposed method. Additionally, our framework captures a disentangled representation of the indirectly observed mediator and yields accurate estimation of the direct and mediated effects in synthetic and semi-synthetic experiments, providing evidence of its potential utility in real-world applications.

Style Transfer Enabled Sim2Real Framework for Efficient Learning of Robotic Ultrasound Image Analysis Using Simulated Data

Robotic ultrasound (US) systems have shown great potential to make US examinations easier and more accurate. Recently, various machine learning techniques have been proposed to realize automatic US image interpretation for robotic US acquisition tasks. However, obtaining large amounts of real US imaging data for training is usually expensive or even unfeasible in some clinical applications. An alternative is to build a simulator to generate synthetic US data for training, but the differences between simulated and real US images may result in poor model performance. This work presents a Sim2Real framework to efficiently learn robotic US image analysis tasks based only on simulated data for real-world deployment. A style transfer module is proposed based on unsupervised contrastive learning and used as a preprocessing step to convert the real US images into the simulation style. Thereafter, a task-relevant model is designed to combine CNNs with vision transformers to generate the task-dependent prediction with improved generalization ability. We demonstrate the effectiveness of our method in an image regression task to predict the probe position based on US images in robotic transesophageal echocardiography (TEE). Our results show that using only simulated US data and a small amount of unlabelled real data for training, our method can achieve comparable performance to semi-supervised and fully supervised learning methods. Moreover, the effectiveness of our previously proposed CT-based US image simulation method is also indirectly confirmed.

Domain Adaptation via Rebalanced Sub-domain Alignment

Unsupervised domain adaptation (UDA) is a technique used to transfer knowledge from a labeled source domain to a different but related unlabeled target domain. While many UDA methods have shown success in the past, they often assume that the source and target domains must have identical class label distributions, which can limit their effectiveness in real-world scenarios. To address this limitation, we propose a novel generalization bound that reweights source classification error by aligning source and target sub-domains. We prove that our proposed generalization bound is at least as strong as existing bounds under realistic assumptions, and we empirically show that it is much stronger on real-world data. We then propose an algorithm to minimize this novel generalization bound. We demonstrate by numerical experiments that this approach improves performance in shifted class distribution scenarios compared to state-of-the-art methods.

Estimating Causal Effects using a Multi-task Deep Ensemble

Over the past few decades, a number of methods have been proposed for causal effect estimation, yet few have been demonstrated to be effective in handling data with complex structures, such as images. To fill this gap, we propose a Causal Multi-task Deep Ensemble (CMDE) framework to learn both shared and group-specific information from the study population and prove its equivalence to a multi-task Gaussian process (GP) with coregionalization kernel a priori. Compared to multi-task GP, CMDE efficiently handles high-dimensional and multi-modal covariates and provides pointwise uncertainty estimates of causal effects. We evaluate our method across various types of datasets and tasks and find that CMDE outperforms state-of-the-art methods on a majority of these tasks.