arXiv subjects

Hongyu Zhou

Publications and source records attributed to Hongyu Zhou.

At least 37 records · Page 2Linked to original sources

arXiv2026

Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters

We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency. We focus on what matters most when building agents: sharp reasoning and fast, reliable execution. Step 3.5 Flash pairs a 196B-parameter foundation with 11B active parameters for efficient inference. It is optimized with interleaved 3:1 sliding-window/full attention and Multi-Token Prediction (MTP-3) to reduce the latency and cost of multi-round agentic interactions. To reach frontier-level intelligence, we design a scalable reinforcement learning framework that combines verifiable signals with preference feedback, while remaining stable under large-scale off-policy training, enabling consistent self-improvement across mathematics, code, and tool use. Step 3.5 Flash demonstrates strong performance across agent, coding, and math tasks, achieving 85.4% on IMO-AnswerBench, 86.4% on LiveCodeBench-v6 (2024.08-2025.05), 88.2% on tau2-Bench, 69.0% on BrowseComp (with context management), and 51.0% on Terminal-Bench 2.0, comparable to frontier models such as GPT-5.2 xHigh and Gemini 3.0 Pro. By redefining the efficiency frontier, Step 3.5 Flash provides a high-density foundation for deploying sophisticated agents in real-world industrial environments.

cs.CL↗

arXiv2026

The Constant Eye: Benchmarking and Bridging Appearance Robustness in Autonomous Driving

Despite rapid progress, autonomous driving algorithms remain notoriously fragile under Out-of-Distribution (OOD) conditions. We identify a critical decoupling failure in current research: the lack of distinction between appearance-based shifts, such as weather and lighting, and structural scene changes. This leaves a fundamental question unanswered: Is the planner failing because of complex road geometry, or simply because it is raining? To resolve this, we establish navdream, a high-fidelity robustness benchmark leveraging generative pixel-aligned style transfer. By creating a visual stress test with negligible geometric deviation, we isolate the impact of appearance on driving performance. Our evaluation reveals that existing planning algorithms often show significant degradation under OOD appearance conditions, even when the underlying scene structure remains consistent. To bridge this gap, we propose a universal perception interface leveraging a frozen visual foundation model (DINOv3). By extracting appearance-invariant features as a stable interface for the planner, we achieve exceptional zero-shot generalization across diverse planning paradigms, including regression-based, diffusion-based, and scoring-based models. Our plug-and-play solution maintains consistent performance across extreme appearance shifts without requiring further fine-tuning. The benchmark and code will be made available.

cs.CV↗

arXiv2026

ReRoPE: Repurposing RoPE for Relative Camera Control

Video generation with controllable camera viewpoints is essential for applications such as interactive content creation, gaming, and simulation. Existing methods typically adapt pre-trained video models using camera poses relative to a fixed reference, e.g., the first frame. However, these encodings lack shift-invariance, often leading to poor generalization and accumulated drift. While relative camera pose embeddings defined between arbitrary view pairs offer a more robust alternative, integrating them into pre-trained video diffusion models without prohibitive training costs or architectural changes remains challenging. We introduce ReRoPE, a plug-and-play framework that incorporates relative camera information into pre-trained video diffusion models without compromising their generation capability. Our approach is based on the insight that Rotary Positional Embeddings (RoPE) in existing models underutilize their full spectral bandwidth, particularly in the low-frequency components. By seamlessly injecting relative camera pose information into these underutilized bands, ReRoPE achieves precise control while preserving strong pre-trained generative priors. We evaluate our method on both image-to-video (I2V) and video-to-video (V2V) tasks in terms of camera control accuracy and visual fidelity. Our results demonstrate that ReRoPE offers a training-efficient path toward controllable, high-fidelity video generation. See project page for more results: https://sisyphe-lee.github.io/ReRoPE/

cs.CV↗

arXiv2026

SRL Proxemics: Spatial Guidelines for Supernumerary Robotic Limbs in Near-Body Interactions

Wearable supernumerary robotic limbs (SRLs) sit at the intersection of human augmentation and embodied AI, transforming into extensions of the human body. However, their movements within the intimate near-body space raise unresolved challenges for perceived safety, user control, and trust. In this paper, we present results from a Wizard-of-Oz study (n=18), where participants completed near-body collaboration tasks with SRLs to explore these challenges. We collected qualitative data through think-aloud protocols and semi-structured interviews, complemented by physiological signals and post-task ratings. Findings indicate that greater autonomy did not inherently enhance perceived safety or trust. Instead, participants identified near-body zones and paired them with clear coordination rules. They also expressed expectations for how different arm components should behave, shaping preferences around autonomy, perceived safety, and trust. Building on these insights, we introduce SRL Proxemics, a zone- and segment-level design framework showing that autonomy is not monolithic: perceived safety hinges on spatially calibrated, legible behaviors, not higher autonomy.

cs.HC↗

arXiv2026

One Body, Two Minds: Alternating VR Perspective During Remote Teleoperation of Supernumerary Limbs

Remote VR teleoperation with supernumerary robotic limbs enables distant users to operate in another's local space. While a shared first-person view aids hand-eye coordination, locking the guest's camera to the host's head can degrade comfort, embodiment, and coordination. Based on a formative study (N=10) using a virtual supernumerary robotic limbs configuration to stress-test coordination, we propose guest-driven perspective switching from a shared first-person baseline (Shared Embodied View) to two alternatives: (a) a stabilized view with guest-controlled rotation (Embedded Anchored View), and (b) a fully decoupled third-person view (Out-of-body View). We ran a user study with 24 pairs (N=48) who switched between the baseline and proposed views as task demands changed. We measured performance, embodiment, fatigue, physiological arousal, and switching behaviors. Our results reveal role-dependent trade-offs: Out-of-body View improves navigation efficiency and reduces errors, while Embedded Anchored View supports embodiment. We conclude with guidelines: use Embedded Anchored View for hand-centric adjustments, Out-of-body View for navigation and object placement, and ensure smooth transitions.

cs.HC↗

arXiv2026

FreeFix: Boosting 3D Gaussian Splatting via Fine-Tuning-Free Diffusion Models

Neural Radiance Fields and 3D Gaussian Splatting have advanced novel view synthesis, yet still rely on dense inputs and often degrade at extrapolated views. Recent approaches leverage generative models, such as diffusion models, to provide additional supervision, but face a trade-off between generalization and fidelity: fine-tuning diffusion models for artifact removal improves fidelity but risks overfitting, while fine-tuning-free methods preserve generalization but often yield lower fidelity. We introduce FreeFix, a fine-tuning-free approach that pushes the boundary of this trade-off by enhancing extrapolated rendering with pretrained image diffusion models. We present an interleaved 2D-3D refinement strategy, showing that image diffusion models can be leveraged for consistent refinement without relying on costly video diffusion models. Furthermore, we take a closer look at the guidance signal for 2D refinement and propose a per-pixel confidence mask to identify uncertain regions for targeted improvement. Experiments across multiple datasets show that FreeFix improves multi-frame consistency and achieves performance comparable to or surpassing fine-tuning-based methods, while retaining strong generalization ability.

cs.CV↗

arXiv2026

STEP3-VL-10B Technical Report

We present STEP3-VL-10B, a lightweight open-source foundation model designed to redefine the trade-off between compact efficiency and frontier-level multimodal intelligence. STEP3-VL-10B is realized through two strategic shifts: first, a unified, fully unfrozen pre-training strategy on 1.2T multimodal tokens that integrates a language-aligned Perception Encoder with a Qwen3-8B decoder to establish intrinsic vision-language synergy; and second, a scaled post-training pipeline featuring over 1k iterations of reinforcement learning. Crucially, we implement Parallel Coordinated Reasoning (PaCoRe) to scale test-time compute, allocating resources to scalable perceptual reasoning that explores and synthesizes diverse visual hypotheses. Consequently, despite its compact 10B footprint, STEP3-VL-10B rivals or surpasses models 10$\times$-20$\times$ larger (e.g., GLM-4.6V-106B, Qwen3-VL-235B) and top-tier proprietary flagships like Gemini 2.5 Pro and Seed-1.5-VL. Delivering best-in-class performance, it records 92.2% on MMBench and 80.11% on MMMU, while excelling in complex reasoning with 94.43% on AIME2025 and 75.95% on MathVision. We release the full model suite to provide the community with a powerful, efficient, and reproducible baseline.

cs.CV↗

arXiv2026

PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning

We introduce Parallel Coordinated Reasoning (PaCoRe), a training-and-inference framework designed to overcome a central limitation of contemporary language models: their inability to scale test-time compute (TTC) far beyond sequential reasoning under a fixed context window. PaCoRe departs from the traditional sequential paradigm by driving TTC through massive parallel exploration coordinated via a message-passing architecture in multiple rounds. Each round launches many parallel reasoning trajectories, compacts their findings into context-bounded messages, and synthesizes these messages to guide the next round and ultimately produce the final answer. Trained end-to-end with large-scale, outcome-based reinforcement learning, the model masters the synthesis abilities required by PaCoRe and scales to multi-million-token effective TTC without exceeding context limits. The approach yields strong improvements across diverse domains, and notably pushes reasoning beyond frontier systems in mathematics: an 8B model reaches 94.5% on HMMT 2025, surpassing GPT-5's 93.2% by scaling effective TTC to roughly two million tokens. We open-source model checkpoints, training data, and the full inference pipeline to accelerate follow-up work.

cs.LG↗

arXiv2026

Feature-based Inversion of 2.5D Controlled Source Electromagnetic Data using Generative Priors

In this study, we investigate feature-based 2.5D controlled source marine electromagnetic (mCSEM) data inversion using generative priors. Two-and-half dimensional modeling using finite difference method (FDM) is adopted to compute the response of horizontal electric dipole (HED) excitation. Rather than using a neural network to approximate the entire inverse mapping in a black-box manner, we adopt a plug-andplay strategy in which a variational autoencoder (VAE) is used solely to learn prior information on conductivity distributions. During the inversion process, the conductivity model is iteratively updated using the Gauss Newton method, while the model space is constrained by projections onto the learned VAE decoder. This framework preserves explicit control over data misfit and enables flexible adaptation to different survey configurations. Numerical and field experiments demonstrate that the proposed approach effectively incorporates prior information, improves reconstruction accuracy, and exhibits good generalization performance.

physics.geo-ph↗

arXiv2025

No-Regret Model Predictive Control with Online Learning of Koopman Operators

We study a problem of simultaneous system identification and model predictive control of nonlinear systems. Particularly, we provide an algorithm for systems with unknown residual dynamics that can be expressed by Koopman operators. Such residual dynamics can model external disturbances and modeling errors, such as wind and wave disturbances to aerial and marine vehicles, or inaccurate model parameters. The algorithm has finite-time near-optimality guarantees and asymptotically converges to the optimal non-causal controller. Specifically, the algorithm enjoys sublinear \textit{dynamic regret}, defined herein as the suboptimality against an optimal clairvoyant controller that knows how the unknown dynamics will adapt to its states and actions. To this end, we assume the algorithm is given Koopman observable functions such that the unknown dynamics can be approximated by a linear dynamical system. Then, it employs model predictive control based on the current learned model of the unknown residual dynamics. This model is updated online using least squares in a self-supervised manner based on the data collected while controlling the system. We validate our algorithm in physics-based simulations of a cart-pole system aiming to maintain the pole upright despite inaccurate model parameters.

eess.SY↗

arXiv2025

Adaptive Legged Locomotion via Online Learning for Model Predictive Control

We provide an algorithm for adaptive legged locomotion via online learning and model predictive control. The algorithm is composed of two interacting modules: model predictive control (MPC) and online learning of residual dynamics. The residual dynamics can represent modeling errors and external disturbances. We are motivated by the future of autonomy where quadrupeds will autonomously perform complex tasks despite real-world unknown uncertainty, such as unknown payload and uneven terrains. The algorithm uses random Fourier features to approximate the residual dynamics in reproducing kernel Hilbert spaces. Then, it employs MPC based on the current learned model of the residual dynamics. The model is updated online in a self-supervised manner using least squares based on the data collected while controlling the quadruped. The algorithm enjoys sublinear \textit{dynamic regret}, defined as the suboptimality against an optimal clairvoyant controller that knows how the residual dynamics. We validate our algorithm in Gazebo and MuJoCo simulations, where the quadruped aims to track reference trajectories. The Gazebo simulations include constant unknown external forces up to $12\boldsymbol{g}$, where $\boldsymbol{g}$ is the gravity vector, in flat terrain, slope terrain with $20\degree$ inclination, and rough terrain with $0.25m$ height variation. The MuJoCo simulations include time-varying unknown disturbances with payload up to $8~kg$ and time-varying ground friction coefficients in flat terrain.

cs.RO↗

arXiv2025

Simultaneous System Identification and Model Predictive Control with No Dynamic Regret

We provide an algorithm for the simultaneous system identification and model predictive control of nonlinear systems. The algorithm has finite-time near-optimality guarantees and asymptotically converges to the optimal (non-causal) controller. Particularly, the algorithm enjoys sublinear dynamic regret, defined herein as the suboptimality against an optimal clairvoyant controller that knows how the unknown disturbances and system dynamics will adapt to its actions. The algorithm is self-supervised and applies to control-affine systems with unknown dynamics and disturbances that can be expressed in reproducing kernel Hilbert spaces. Such spaces can model external disturbances and modeling errors that can even be adaptive to the system's state and control input. For example, they can model wind and wave disturbances to aerial and marine vehicles, or inaccurate model parameters such as inertia of mechanical systems. The algorithm first generates random Fourier features that are used to approximate the unknown dynamics or disturbances. Then, it employs model predictive control based on the current learned model of the unknown dynamics (or disturbances). The model of the unknown dynamics is updated online using least squares based on the data collected while controlling the system. We validate our algorithm in both hardware experiments and physics-based simulations. The simulations include (i) a cart-pole aiming to maintain the pole upright despite inaccurate model parameters, and (ii) a quadrotor aiming to track reference trajectories despite unmodeled aerodynamic drag effects. The hardware experiments include a quadrotor aiming to track a circular trajectory despite unmodeled aerodynamic drag effects, ground effects, and wind disturbances.

cs.RO↗

arXiv2025

Hazard-Responsive Digital Twin for Climate-Driven Urban Resilience and Equity

Compounding climate hazards, such as wildfire-induced outages and urban heatwaves, challenge the stability and equity of cities. We present a Hazard-Responsive Digital Twin (H-RDT) that combines physics-informed neural network modeling, multimodal data fusion, and equity-aware risk analytics for urban-scale response. In a synthetic district with diverse building archetypes and populations, a simulated wildfire-outage-heatwave cascade shows that H-RDT maintains stable indoor temperature predictions (approximately 31 to 33 C) under partial sensor loss, reproducing outage-driven surges and recovery. The reinforcement learning based fusion module adaptively reweights IoT, UAV, and satellite inputs to sustain spatiotemporal coverage, while the equity-adjusted mapping isolates high-vulnerability clusters (schools, clinics, low-income housing). Prospective interventions, such as preemptive cooling-center activation and microgrid sharing, reduce population-weighted thermal risk by 11 to 13 percent, shrink the 95th-percentile (tail) risk by 7 to 17 percent, and cut overheating hours by up to 9 percent. Beyond the synthetic demonstration, the framework establishes a transferable foundation for real-city implementation, linking physical hazard modeling with social equity and decision intelligence. The H-RDT advances digital urban resilience toward adaptive, learning-based, and equity-centered decision support for climate adaptation.

cs.LG↗

arXiv2025

Towards Depth Foundation Model: Recent Trends in Vision-Based Depth Estimation

Depth estimation is a fundamental task in 3D computer vision, crucial for applications such as 3D reconstruction, free-viewpoint rendering, robotics, autonomous driving, and AR/VR technologies. Traditional methods relying on hardware sensors like LiDAR are often limited by high costs, low resolution, and environmental sensitivity, limiting their applicability in real-world scenarios. Recent advances in vision-based methods offer a promising alternative, yet they face challenges in generalization and stability due to either the low-capacity model architectures or the reliance on domain-specific and small-scale datasets. The emergence of scaling laws and foundation models in other domains has inspired the development of "depth foundation models": deep neural networks trained on large datasets with strong zero-shot generalization capabilities. This paper surveys the evolution of deep learning architectures and paradigms for depth estimation across the monocular, stereo, multi-view, and monocular video settings. We explore the potential of these models to address existing challenges and provide a comprehensive overview of large-scale datasets that can facilitate their development. By identifying key architectures and training strategies, we aim to highlight the path towards robust depth foundation models, offering insights into their future research and applications.

cs.CV↗

arXiv2025

Arab Spring's Impact on Science through the Lens of Scholarly Attention, Funding, and Migration

The 2010-2011 Arab Spring reverberated far beyond politics, reshaping how the Middle East and North Africa region (MENA) is studied. Analyzing 3.7 million Scopus-indexed articles published between 2002 and 2019, we find that mentions of ten of these countries in titles or abstracts rose significantly after 2011 relative to the global baseline, with Egypt receiving the greatest attention in the region. We link this surge to two intertwined mechanisms: an increase in research funding directed at the MENA region and the emigration of researchers who continued publishing on their countries of origin. Our analysis reveals that Saudi Arabia has emerged as a regional hub for studying the affected countries, attracting funding and scholars, and thereby playing a significant role in shaping the scientific narrative on the region. These findings demonstrate how political upheaval can reshape global knowledge flows by altering who studies whom, with what resources, and in which disciplines.

cs.DL↗

arXiv2025

JRN-Geo: A Joint Perception Network based on RGB and Normal images for Cross-view Geo-localization

Cross-view geo-localization plays a critical role in Unmanned Aerial Vehicle (UAV) localization and navigation. However, significant challenges arise from the drastic viewpoint differences and appearance variations between images. Existing methods predominantly rely on semantic features from RGB images, often neglecting the importance of spatial structural information in capturing viewpoint-invariant features. To address this issue, we incorporate geometric structural information from normal images and introduce a Joint perception network to integrate RGB and Normal images (JRN-Geo). Our approach utilizes a dual-branch feature extraction framework, leveraging a Difference-Aware Fusion Module (DAFM) and Joint-Constrained Interaction Aggregation (JCIA) strategy to enable deep fusion and joint-constrained semantic and structural information representation. Furthermore, we propose a 3D geographic augmentation technique to generate potential viewpoint variation samples, enhancing the network's ability to learn viewpoint-invariant features. Extensive experiments on the University-1652 and SUES-200 datasets validate the robustness of our method against complex viewpoint ariations, achieving state-of-the-art performance.

cs.CV↗

arXiv2025

Step-Audio 2 Technical Report

This paper presents Step-Audio 2, an end-to-end multi-modal large language model designed for industry-strength audio understanding and speech conversation. By integrating a latent audio encoder and reasoning-centric reinforcement learning (RL), Step-Audio 2 achieves promising performance in automatic speech recognition (ASR) and audio understanding. To facilitate genuine end-to-end speech conversation, Step-Audio 2 incorporates the generation of discrete audio tokens into language modeling, significantly enhancing its responsiveness to paralinguistic information such as speaking styles and emotions. To effectively leverage the rich textual and acoustic knowledge in real-world data, Step-Audio 2 integrates retrieval-augmented generation (RAG) and is able to call external tools such as web search to mitigate hallucination and audio search to switch timbres. Trained on millions of hours of speech and audio data, Step-Audio 2 delivers intelligence and expressiveness across diverse conversational scenarios. Evaluation results demonstrate that Step-Audio 2 achieves state-of-the-art performance on various audio understanding and conversational benchmarks compared to other open-source and commercial solutions. Please visit https://github.com/stepfun-ai/Step-Audio2 for more information.

cs.CL↗

arXiv2025

NextStep-1: Toward Autoregressive Image Generation with Continuous Tokens at Scale

Prevailing autoregressive (AR) models for text-to-image generation either rely on heavy, computationally-intensive diffusion models to process continuous image tokens, or employ vector quantization (VQ) to obtain discrete tokens with quantization loss. In this paper, we push the autoregressive paradigm forward with NextStep-1, a 14B autoregressive model paired with a 157M flow matching head, training on discrete text tokens and continuous image tokens with next-token prediction objectives. NextStep-1 achieves state-of-the-art performance for autoregressive models in text-to-image generation tasks, exhibiting strong capabilities in high-fidelity image synthesis. Furthermore, our method shows strong performance in image editing, highlighting the power and versatility of our unified approach. To facilitate open research, we will release our code and models to the community.

cs.CV↗