CVPR 2026 论文列表

Retrieval-Augmented Generation (RAG) has recently been extended to multimodal settings, connecting multimodal large language models (MLLMs) with vast corpora of external knowledge such as multimodal knowledge graphs (MMKGs). Despite their recent success, multimodal RAG in the audio-visual domain remains challenging due to 1) limited modality coverage and multi-hop connectivity of existing MMKGs, and 2) retrieval based solely on similarity in a shared multimodal embedding space, which fails to filter out off-topic or redundant knowledge. To address these limitations, we propose M^3KG-RAG, a Multi-hop Multimodal Knowledge Graph-enhanced RAG that retrieves query-aligned audio-visual knowledge from MMKGs, improving reasoning depth and answer faithfulness in MLLMs. Specifically, we devise a lightweight multi-agent pipeline to construct multi-hop MMKG (M^3KG), which contains context-enriched triplets of multimodal entities, enabling modality-wise retrieval based on input queries. Furthermore, we introduce GRASP (Grounded Retrieval And Selective Pruning), which ensures precise entity grounding to the query, evaluates answer-supporting relevance, and prunes redundant context to retain only knowledge essential for response generation. Extensive experiments across diverse multimodal benchmarks demonstrate that M^3KG-RAG significantly enhances MLLMs' multimodal reasoning and grounding over existing approaches.

Adapting production-level computer vision tools to bespoke scientific datasets is a critical "last mile" bottleneck. Current solutions are impractical: fine-tuning requires large annotated datasets scientists often lack, while manual code adaptation costs scientists weeks to months of effort. We consider using AI agents to automate this manual coding, and focus on the open question of optimal agent design for this targeted task. We introduce a systematic evaluation framework for agentic code optimization and use it to study three production-level biomedical imaging pipelines. We demonstrate that a simple agent framework consistently generates adaptation code that outperforms human-expert solutions. Our analysis reveals that common, complex agent architectures are not universally beneficial, leading to a practical roadmap for agent design. We open source our framework and validate our approach by deploying agent-generated functions into a production pipeline, demonstrating a clear pathway for real-world impact.

Human-level agentic intelligence extends beyond low-level geometric perception, evolving from recognizing where things are to understanding what they are for. While existing benchmarks effectively evaluate the geometric perception capabilities of multimodal large language models (MLLMs), they fall short of probing the higher-order cognitive abilities required for grounded intelligence. To address this gap, we introduce the Spatial-Functional Intelligence Benchmark (SFI-Bench), a video-based benchmark with over 1,500 expert-annotated questions derived from diverse egocentric indoor video scans. SFI-Bench systematically evaluates two complementary dimensions of advanced reasoning: (1) Structured Spatial Reasoning, which requires understanding complex layouts and forming coherent spatial representations, and (2) Functional Reasoning, which involves inferring object affordances and their context-dependent utility. The benchmark includes tasks such as conditional counting, multi-hop relational reasoning, functional pairing, and knowledge-grounded troubleshooting, directly challenging models to integrate perception, memory, and inference. Our experiments reveal that current MLLMs consistently struggle to combine spatial memory with functional reasoning and external knowledge, highlighting a critical bottleneck in achieving grounded intelligence. SFI-Bench therefore provides a diagnostic tool for measuring progress toward more cognitively capable and truly grounded multimodal agents.

Vision Language Models (VLMs) demonstrate strong qualitative visual understanding, but struggle with metrically precise spatial reasoning required for embodied applications. The agentic paradigm promises that VLMs can use a wide variety of tools that could augment these capabilities, such as depth estimators, segmentation models, and pose estimators. Yet it remains an open challenge how to realize this vision without solely relying on handcrafted prompting strategies or enforcing fixed, predefined tool pipelines that limit VLMs' ability to discover optimal tool-use patterns. Reinforcement Learning could overcome this gap, but has so far been limited to reasoning with a single visual tool due to the large search space in multi-tool reasoning. We introduce Double Interactive Reinforcement Learning (DIRL), a two-phase training framework where VLMs learn to coordinate multiple tools through interactive exploration and feedback. In the teaching phase, we combine demonstrations from a single tool specialist trained via interactive RL with traces from a frontier model using all tools. In the exploration phase, the model further refines multi-tool coordination through continued RL. Our model, SpaceTools, with tool-augmented spatial reasoning ability, achieves state-of-the-art performance on spatial understanding benchmarks (RoboSpatial-Home, BLINK, Internal Benchmark) and demonstrates reliable real-world manipulation using a 7-DOF robot as a tool. DIRL provides substantial improvements over the vanilla SFT (+12% on RoboSpatial) and RL (+16% on RoboSpatial) baselines. Project page: https://spacetools.github.io/.

Real-world data collection for embodied agents remains costly and unsafe, calling for scalable, realistic, and simulator-ready 3D environments. However, existing scene-generation systems often rely on rule-based or task-specific pipelines, yielding artifacts and physically invalid scenes. We present SAGE, an agentic framework that, given a user-specified embodied task (e.g., "pick up a bowl and place it on the table"), understands the intent and automatically generates simulation-ready environments at scale. The agent couples multiple generators for layout and object composition with critics that evaluate semantic plausibility, visual realism, and physical stability. Through iterative reasoning and adaptive tool selection, it self-refines the scenes until meeting user intent and physical validity. The resulting environments are realistic, diverse, and directly deployable in modern simulators for policy training. Policies trained purely on this data exhibit clear scaling trends and generalize to unseen objects and layouts, demonstrating the promise of simulation-driven scaling for embodied AI. Code, demos, and the SAGE-10k dataset can be found on the project page: https://research.nvidia.com/labs/dir/sage/.

Humanoid robotics has strong potential to transform daily service and caregiving applications. Although recent advances in general motion tracking within physics engines (GMT) have enabled virtual characters and humanoid robots to reproduce a broad range of human motions, these behaviors are primarily limited to contact-less social interactions or isolated movements. Assistive scenarios, by contrast, require continuous awareness of a human partner and rapid adaptation to their evolving posture and dynamics. In this paper, we formulate the imitation of closely interacting, force-exchanging human-human motion sequences as a multi-agent reinforcement learning problem. We jointly train partner-aware policies for both the supporter (assistant) agent and the recipient agent in a physics simulator to track assistive motion references. To make this problem tractable, we introduce a partner policies initialization scheme that transfers priors from single-human motion-tracking controllers, greatly improving exploration. We further propose dynamic reference retargeting and contact-promoting reward, which adapt the assistant's reference motion to the recipient's real-time pose and encourage physically meaningful support. We show that \model is the first method capable of successfully tracking assistive interaction motions on established benchmarks, demonstrating the benefits of a multi-agent RL formulation for physically grounded and socially aware humanoid control.

Foley art plays a pivotal role in enhancing immersive auditory experiences in film, yet manual creation of spatio-temporal aligned audio remains labor-intensive. We propose FoleyDesigner, a novel framework inspired by professional Foley workflows, integrating film clip analysis, spatio-temporal controllable Foley generation, and professional mixing capabilities. Technically, FoleyDesigner employs a multi-agent architecture for precise spatio-temporal analysis. It achieves spatio-temporal alignment through latent diffusion models trained on spatiotemporal cues extracted from video frames, combined with large language model (LLM)-driven hybrid mechanisms that emulate film industry-grade post-production practices. To address the lack of high-quality stere Foley datasets in film, we introduce FilmStereo, the first professional stereo Foley dataset containing spatial metadata, precise timestamps, and semantic annotations for eight common Foley categories. For application, the framework supports interactive user control while maintaining seamless integration with professional pipelines, including 5.1-channel Dolby Atmos systems compliant with ITU-R BS.775 standards, thereby offering extensive creative flexibility. Extensive experiments demonstrate that our method achieves superior spatio-temporal alignment compared to existing baselines, with integration validated in film industrial-grade workflows.

Recent advances in multimodal large language models (MLLMs) have accelerated progress in domain-oriented AI, yet their development in geoscience and remote sensing (RS) remains constrained by distinctive challenges: wide-ranging disciplinary knowledge, heterogeneous sensor modalities, and a fragmented spectrum of tasks. To bridge these gaps, we introduce GeoMMBench, a comprehensive multimodal question-answering benchmark covering diverse RS disciplines, sensors, and tasks, enabling broader and more rigorous evaluation than prior benchmarks. Using GeoMMBench, we assess 36 open-source and proprietary large language models (LLMs), uncovering systematic deficiencies in domain knowledge, perceptual grounding, and reasoning--capabilities essential for expert-level geospatial interpretation. Beyond evaluation, we propose GeoMMAgent, a multi-agent framework that strategically integrates retrieval, perception, and reasoning through domain-specific RS models and tools. Extensive experimental results demonstrate that GeoMMAgent significantly outperforms standalone LLMs, underscoring the importance of tool-augmented agents for dynamically tackling complex geoscience and RS challenges.

Sketching in 3D space enables expressive reasoning about shape, structure, and spatial relationships, yet generating 3D sketches through natural language remains a major challenge. In this work, we introduce 3DrawAgent, a training-free, language-driven framework for 3D sketch generation that leverages large language models (LLMs) to sequentially draw 3D Bezier curves under geometric feedback. Unlike prior 2D sketch agents, our method introduces a relative experience optimization strategy that adapts the recently proposed Group Reward Policy Optimization (GRPO) paradigm. Instead of relying on explicit ground-truth supervision, we construct pairwise comparisons among generated sketches, with each pair consisting of a relatively better and a worse result based on CLIP-based perceptual rewards and LLM-based fine-grained qualitative assessment. These experiences are then used to iteratively refine the prior knowledge of 3D drawing, enabling black-box reinforcement of the model's 3D awareness. This design allows our model to self-improve its spatial understanding and drawing quality without parameter updates. Experiments show that 3DrawAgent can generate complex and coherent 3D Bezier sketches from diverse textual prompts, exhibit emergent geometric reasoning, and generalize to novel shapes, establishing a new paradigm for advancing the field of training-free 3D sketch intelligence.

Reward models are critical for aligning vision-language systems with human preferences, yet current approaches suffer from hallucination, weak visual grounding, and an inability to use tools for verification, limiting their reliability on complex multimodal reasoning tasks.We present **ARM-Thinker**, an **A**gentic multimodal **R**eward **M**odel that autonomously invokes external tools (e.g., image cropping, doc page retrieval) to ground judgments in verifiable evidence, replacing static, non-interactive reward scoring.This enables the model to verify fine-grained visual details, cross-reference multi-page evidence, and validate reasoning claims, which are capabilities absent in existing reward models.We train ARM-Thinker with multi-stage reinforcement learning, jointly optimizing tool-calling decisions and judgment accuracy.To evaluate agentic reward modeling, we introduce **ARMBench-VL**, comprising three benchmarks that assess fine-grained visual grounding (image-level tools), multi-page document understanding (retrieval tools), and instruction following (text-level verification).ARM-Thinker achieves +16.2% average improvement on reward modeling benchmarks, +9.6% on tool-use tasks, and outperforms baselines on multimodal math and logical reasoning benchmarks.Our results demonstrate that agentic capabilities significantly enhance both accuracy and interpretability of reward models.

Human communication is inherently multimodal and social: words, prosody, and body language jointly carry intent. Yet most prior systems model human behavior as a translation task--co-speech gesture or text-to-motion that maps a fixed utterance to motion clips--without requiring agentic decision-making about when to move, what to do, or how to adapt across multi-turn dialogue. This leads to brittle timing, weak social grounding, and fragmented stacks where speech, text, and motion are trained or inferred in isolation. We introduce ViBES (Voice in Behavioral Expression and Synchrony), a conversational 3D agent that jointly plans language and movement and executes dialogue-conditioned body actions. Concretely, ViBES is a speech-language-behavior (SLB) model with a mixture-of-modality-experts (MoME) backbone: modality-partitioned transformer experts for speech, facial expression, and body motion. The model processes interleaved multimodal token streams with hard routing by modality (parameters are split per expert), while sharing information through cross-expert attention. By leveraging strong pretrained speech-language models, the agent supports mixed-initiative interaction: users can speak, type, or issue body-action directives mid-conversation, and the system exposes controllable behavior hooks for streaming responses. We further benchmark on multi-turn conversation with automatic metrics of dialogue-motion alignment and behavior quality, and observe consistent gains over strong co-speech and text-to-motion baselines. ViBES goes beyond "speech-conditioned motion generation" toward agentic virtual bodies where language, prosody, and movement are jointly generated, enabling controllable, socially competent 3D interaction.

Recent advancements in Vision Language Models (VLMs) have expanded their capabilities to interactive agent tasks, yet existing benchmarks remain limited to single-agent or text-only environments. In contrast, real-world scenarios often involve multiple agents interacting within rich visual and textual contexts, posing challenges with both multimodal observations and strategic interactions. To bridge this gap, we introduce Visual Strategic Bench (VS-Bench), a multimodal benchmark that evaluates VLMs for strategic abilities in multi-agent environments. VS-Bench comprises ten vision-grounded environments that cover cooperative, competitive, and mixed-motive interactions. The performance of VLM agents is evaluated across three dimensions: perception measured by element recognition accuracy; strategic reasoning measured by next-action prediction accuracy; and decision-making measured by normalized episode return. Extensive experiments on fifteen leading VLMs show that, although current models exhibit strong perception abilities, there remains a significant gap to optimal performance in reasoning and decision-making, with the best-performing model attaining 46.6% prediction accuracy and 31.4% normalized return. We further analyze the key factors influencing performance, conduct human studies, and examine failure modes to provide a deeper understanding of VLMs' strategic abilities. By standardizing the evaluation and highlighting the limitations of existing models, we envision VS-Bench as a foundation for future research on strategic multimodal agents.

Multimodal language models (MLLMs) are increasingly paired with vision tools (e.g., depth, flow, correspondence) to enhance visual reasoning. However, despite access to these tool-generated visual cues, MLLMs often fail to benefit fully from them. Existing approaches typically feed raw tool outputs into the model, but these dense, pixel-level representations are misaligned with the language-native reasoning strengths of LLMs, leading to weak perception due to reliance on language priors. We argue that, in problems where vision tools can provide the necessary visual cues, the bottleneck is not more tool calls or larger MLLMs, it is how tool outputs are represented. We introduce Perception Programs (P^2), a training-free, model-agnostic method that rewrites tool outputs into compact, structured, language-native summaries that MLLMs can directly parse and reason over. Across six perception-centric tasks in BLINK, P^2 consistently yields large improvements over base models and raw tool-augmented baselines. With GPT-5 Mini as the base model, P^2 raises its accuracy from 41.35% to 86.47% on multi-view reasoning, from 52.42 to 81.45 on relative depth, and achieves a 19.66 overall average gain across tasks, setting new state-of-the-art results. Even on smaller MLLMs, e.g., InternVL3.5-4B and Qwen3VL-4B, we observe 21-25 absolute gains from P^2 across tasks when comparing to image-based tool variants, surpassing prior agentic, supervised, and RL-based tool-use methods, without any training or model modifications.

MLLMs exhibit strong reasoning on isolated queries, yet they operate de novo--solving each problem independently and often repeating the same mistakes. Existing memory-augmented agents mainly store past trajectories for reuse. However, trajectory-based memory suffers from brevity bias, gradually losing essential domain knowledge. More critically, even in truly multimodal problem-solving settings, it records only a **single-modality** trace of past behavior, failing to preserve how visual attention and logical reasoning jointly contributed to the solution. This is fundamentally misaligned with human cognition: semantic memory is both **multimodal and integrated**, preserving visual and abstract knowledge through coordinated but distinct representational streams. We thus introduce **ViLoMem**, a dual-stream memory framework that constructs compact, schema-based memory. It separately encodes visual distraction patterns and logical reasoning errors, enabling MLLMs to learn from their successful and failed experiences. Following a grow-and-refine principle, the system incrementally accumulates and updates multimodal semantic knowledge--preserving stable, generalizable strategies while avoiding catastrophic forgetting. Across nine multimodal benchmarks, **ViLoMem** consistently improves pass@1 accuracy and substantially reduces repeated visual and logical errors. Ablations confirm the necessity of dual-stream memory with explicit distraction-hallucination separation, demonstrating the value of error-aware multimodal memory for lifelong and cross-domain agentic learning.

Offline Multi-Agent Reinforcement Learning (MARL) is critical for coordinating multiple agents in costly and unsafe environments, yet existing methods struggle with high sensitivity to reward functions and weak generalization to new goals, limiting its practical impact. Inspired by single-agent Offline Goal-Conditioned RL (OGCRL), we propose the first goal-conditioned offline MARL framework, extending OGCRL to multi-agent settings under both fully decentralized and centralized training with decentralized execution (CTDE) paradigms. To systematically evaluate this setting, we introduce MangoBench, the first fully cooperative multi-goal benchmark for MARL, covering 3 environments, 4 agent types, and 47 tasks, designed to assess joint-control locomotion, synchronous and asynchronous bimanual manipulation, and robustness to high-dimensional inputs. Extensive experiments demonstrate that our baselines achieve strong multi-goal generalization under sparse rewards, yet no method dominates all tasks, revealing both the intrinsic complexity and the unexplored potential of goal-conditioned offline MARL.

Graphical User Interface (GUI) agents require effective utilization of historical context to perform sequential navigation tasks. While incorporating past actions and observations can significantly improve decision-making, naively using full history leads to excessive computational overhead and potential distraction from irrelevant information. In this work, we introduce ****HiconAgent****, a GUI agent trained with ****History Context-aware Policy Optimization (HCPO)**** for effective and efficient utilization of historical information. HCPO explicitly optimizes history usage in both sampling and policy updates by integrating two complementary components: ****(1) Dynamic Context Sampling (DCS)**** presents the agent with variable-length histories during sampling, enabling adaptive use of the most relevant historical context to improve sequential decision quality; ****(2) Anchor-guided History Compression (AHC)**** refines the policy update phase via a dual-branch optimization strategy, where the compressed branch drops history observations while keeping history actions as information flow anchors. The compressed and uncompressed branches are coupled through a history-enhanced alignment loss to enforce consistent history usage, achieving efficiency with minimal performance degradation. Extensive experiments on mainstream GUI navigation benchmarks demonstrate the strong performance of our model. Despite its smaller size, HiconAgent-3B outperforms GUI-R1-7B by ****+8.46****% grounding and ****+11.32****% step successful rate on GUI-Odyssey, while achieving comparable results on AndroidControl and AITW, with up to ****2.47x computational speedup**** and ****60% FLOPs reduction****.

Vision Language Models (VLMs) exhibit a fundamental semantic-to-geometric gap in spatial reasoning: they excel at qualitative semantic inference but their reasoning operates within a lossy semantic space, misaligned with high-fidelity geometry. Current paradigms fail to bridge this gap. Training-based methods suffer from an "oracle paradox," learning flawed spatial logic from imperfect oracles. Tool-integrated methods constrain the final computation but critically leave the VLM's planning process unconstrained, resulting in geometrically flawed plans. In this work, we propose Geometrically-Constrained Agent (GCA), a training-free agentic paradigm that resolves this gap by introducing a formal task constraint. Specifically, we strategically decouples the VLM's role into two stages. First, acting as a semantic analyst, the VLM translates the user's ambiguous query into the formal, verifiable task constraint, which defines the reference frame and objective. Second, acting as a task solver, the VLM generates and executes tool calls strictly within the deterministic bounds defined by the constraint. This geometrically-constrained strategy successfully resolve the semantic-to-geometric gap, yielding a robust and verifiable reasoning pathway for complex spatial tasks. Comprehensive experiments demonstrate that GCA achieves SOTA performance on multiple spatial reasoning benchmarks, surpassing existing training-based and tool-integrated methods by 27%.

Recent progress in Multimodal Large Language Models (MLLMs) has enabled mobile GUI agents capable of visual perception, cross-modal reasoning, and interactive control. However, existing benchmarks are largely English-centric and fail to capture the linguistic and interaction characteristics of the Chinese mobile ecosystem. They also focus on isolated skills such as GUI grounding or offline agent, lacking a unified and fine-grained framework to assess the full capability chain from perception to execution. To address this gap, we introduce GUI-CEval, the first comprehensive benchmark for Chinese mobile GUI agents, built entirely on physical device environments. GUI-CEval spans 201 mainstream apps across four device types and adopts a two-level structure that evaluates both atomic abilities and realistic application-level performance along five dimensions: perception, planning, reflection, execution, and evaluation. All data are collected and verified through multi-stage manual processes to ensure authenticity and reproducibility. Extensive experiments on 20 representative MLLMs and multi-agent systems show that while models such as Qwen2.5-VL and UI-TARS perform competitively, most MLLMs still exhibit clear weaknesses in reflective decision-making and post-action self-evaluation, limiting their reliability in real-world interactions. We hope GUI-CEval provides a comprehensive benchmark to guide capability diagnosis and advance the development of Chinese mobile GUI agents.

Autonomous Graphical User Interface (GUI) agents powered by Multimodal Large Language Models (MLLMs) are increasingly vital for complex task automation. However, their capacity for self-driven decision-making introduces significant, yet underexplored, security risks, among which backdoor attacks pose a particularly stealthy and high-impact threat. Prior work has shown GUI agents vulnerable to such attacks, but existing methods rely on static trigger-action mappings that execute fixed, context-agnostic behaviors, making them highly detectable. To address this limitation, we introduce **AdapAction**, a novel backdoor attack that subverts the agent's decision-making by embedding an **adaptive, context-aware policy**. Unlike traditional approaches, AdapAction enables the agent to autonomously select environmentally coherent malicious actions based on the current GUI state and user instruction, thereby evading detection while preserving functional utility. Extensive experiments on the Android-In-The-Zoo (AitZ) and AndroidControl benchmarks show that AdapAction achieves up to 100% Attack Success Rate (ASR) while preserving benign task utility. More critically, AdapAction consistently evades a multi-principle-based LLM defense evaluating instruction alignment, visual coherence, and safety, whereas traditional fixed-action attacks are easily detected. This resilience stems from AdapAction's contextually grounded malicious actions, which are semantically and visually indistinguishable from legitimate operations. As a result, AdapAction exhibits exceptional stealth and poses a significantly greater real-world threat to LLM-powered GUI agents.