Tsinghua AI

Packed trajectories, not sequential rollouts, unlock high-fidelity multi-turn image editing by preventing visual drift.

Targeted data augmentation and joint RL/OPD training unlock significantly stronger long-horizon multimodal search capabilities for agents.

Continuous, decomposed visual verification unlocks robust rule-governed generative world modeling, outperforming binary critics and enabling structured reasoning agents.

Visual generation supervision injects significant, yet often inaccessible, knowledge into UMMs, requiring novel alignment strategies to unlock its full potential for text-based reasoning.

Open multimodal code intelligence models now rival proprietary systems, driven by a novel rubric-based data synthesis pipeline that precisely evaluates program correctness, code quality, and visual output.

Current chemistry benchmarks conflate memorization with derivation and treat visual/textual inputs interchangeably; ChemArgus finally provides fine-grained diagnostic scores that pinpoint *where* a model's reasoning chain breaks.

Ultralight action heads achieve state-of-the-art embodied task performance by learning to selectively retrieve visual information via shared latent bridge queries.

Irrelevant text distractors cripple mixed-modality retrieval performance more than images, revealing a text modality preference that degrades performance in a V-shaped curve.

Ambiguous object selection in text-guided image editing is solved by directly annotating the canvas, achieving state-of-the-art performance in complex edits.

Computer-use agents can bypass brittle, hand-engineered APIs entirely by learning when to substitute mouse clicks with native shell commands, yielding a 14.8-point leap on OSWorld.

Purely autoregressive VLAs no longer require awkward hybrid diffusion heads to achieve fine-grained robot control: structuring discrete action tokenization as causally annealed flow matching recovers continuous diffusion precision while preserving end-to-end token efficiency.

RoboDawn is investigated, a human-intuitive interface that exposes robotic control to an agentic VLM through a compact set of discrete translation, rotation, and gripper commands, and an in-context learning (ICL) scheme is introduced that ground the VLM in both interface usage and task-solving strategies.

RiskChainBench introduces RiskChainBench, pairing 3,600 synthetic token-text restoration inputs from 600 source sessions with 600 corresponding human-labeled local web environments that produces a frozen, evidence-cited risk report without message-side semantics or domain-reputation cues.

MUUNRiver-Bench is presented, a diagnostic benchmark whose reference-audio queries use natural-language instructions to define relevance, and task-wise rank reversals reveal complementary biases: acoustic encoders favour local identity, whereas text-aligned encoders favour semantic relations.

Findings show that foundation models require CGM-specific adaptation for reliable forecasting and that dietary context provides clinically meaningful signals beyond CGM alone, especially during postprandial periods.

Vidu S2, which comprises Vidu S2-Avatar, a real-time interactive digital-character model, and Vidu S2-Editing, a real-time video editing model, is presented and the feasibility of real-time spatial video generation for both Vidu S2-Avatar and Vidu S2-Editing is explored.

TANGO enables humanoid robots to navigate cluttered spaces with unprecedented language-guided precision, achieving state-of-the-art performance without real-world training.

This work presents the first synergistic framework that tightly couples hierarchical memory with proactive tool invocation in a closed reasoning loop, and reveals the complementary roles of hierarchical memory and tool strategies.

Turning published scientific figures back into executable Python remains a brittle frontier for top VLMs, with even frontier models routinely failing on multi-component layouts, precise axis semantics, and domain typography.

Natural-language instructions can now handle both zero-shot speech synthesis and surgical acoustic editing within a single unified model, operating at 4-step distilled inference speeds without classifier-free guidance.

A controlled pure-autoregressive testbed is built and track task-specific validation losses during multimodal continual pretraining across text, image, text-to-image (T2I), and image-to-text (I2T) prediction, showing that better reconstruction does not necessarily yield lower task-specific losses or stronger downstream performance, and that image tokenizer choice can affect text modeling under joint optimization.

Single vision foundation models inevitably overfit to deepfake artifacts, but dynamically arbitrating between CLIP's semantics and DINO's structural features via prediction entropy delivers state-of-the-art generalization across unseen generators.

Robots can now perform complex manipulation tasks more effectively by understanding and responding to contact forces without additional sensors.

ToPO achieves superior image generation performance by optimizing token-conditioned preferences, outperforming existing methods across multiple evaluation metrics.

Scheduling imagination in VLA models can cut GPU costs by 80% while boosting performance and robustness in real-world tasks.

Visual design no longer requires choosing between diffusion fidelity and code editability: orchestrating modular asset diffusion through a VLM-driven HTML/CSS coding loop delivers fully interactive, layer-decoupled graphic layouts.

SR-Edit achieves unprecedented image editing fidelity by refining region separation and preserving non-edit areas without introducing new artifacts.

ExBind reveals that even high-validity models can struggle with exact accuracy, exposing critical gaps in visual-to-executable correspondence in AI systems.

TUTTI achieves unprecedented A2S performance by leveraging a synthetic dataset, outperforming traditional methods and enabling robust cross-instrument adaptability.

Credit-addressable reasoning boosts multimodal geometry accuracy by over 8 points, revealing the critical role of structured learning in complex tasks.

Image corruptions can drastically impair retrieval performance, revealing that clean image-only queries often outperform joint retrieval in snap-and-ask scenarios.

A staggering 42.5% of correct answers in long-document VQA can't be derived from terminal working memory alone, revealing a crucial oversight in current evaluation methods.

Reducing visual token exposure by leveraging selective retrieval can dramatically enhance the performance of 3D medical image question answering systems.

Achieving a 25% reduction in prediction error, MOSH-WM redefines the landscape of object-centric video forecasting by grounding state representations in visual support.

Typed spatial readouts can elevate robotic execution success rates by over 50% while significantly reducing processing time.

Object-Uni transforms how we understand and generate spatial representations of objects, enabling precise manipulation of their poses in generated images.

"Prompt-form collapse" can drastically reduce success rates, but TOWN-VLA's controlled intervention boosts performance by ensuring only meaningful prompts are executed.

Emotion-aware art generation just got a major upgrade with ReART, achieving perfect alignment with target emotions through structured visual retrieval.

RuleMaze reveals that separating perception, execution, and rule verification can dramatically enhance MLLMs' ability to follow complex natural-language instructions in spatial planning tasks.

DECOWAM achieves a 21.7% reduction in action prediction error while maintaining robust task performance, showcasing the power of embodiment-aware factorization in mobile manipulation.

By combining images with text queries, ID-VTG significantly improves the accuracy of video grounding in scenarios with visually similar entities.

MedUAG sets a new standard in medical multimodal models, achieving strong performance across diverse understanding and generation tasks with the largest dataset yet.

Policies may succeed in tasks but still violate deformation tolerances, revealing a critical gap in current evaluation methods for deformable-object manipulation.

MSEditor achieves unprecedented consistency in multi-shot video editing, outperforming existing methods by effectively managing identity drift and temporal coherence.

GeoWeaver achieves unprecedented accuracy in long-sequence 3D reconstruction by correcting accumulated errors through a novel combination of geometric priors and adaptive refinement techniques.

Achieving the best macro-average results on dynamic scene reconstruction benchmarks, UniQuery4R redefines efficiency in 4D scene understanding.

Selectively routed stereo evidence boosts humanoid VLA control success rates, achieving 100% grasp success even under severe occlusion.

A non-invasive framework that enhances pretrained generative policies with force control, achieving remarkable improvements in contact stability and execution speed.