双榜领先
01 具身导航
Embodied Navigation
榜单荣誉：两大权威评测集 R2R、RxR 取得领先地位
核心突破
z0
全球首个持续自主进化的物理智能体
性能领先全球新标杆
z0 产品在多个世界级权威评测集中均登顶领先，重新定义具身智能全球新标杆
双榜领先
02 流式视频理解
Streaming Video Understanding
榜单荣誉：两大流式基准 Streamingbench、OVO-Bench 取得领先地位
核心突破
Streamingbench（2 FPS）
准确率Accuracy(%)OVO-Bench
准确率Accuracy(%)三榜领先
03 智能体记忆
Agent Memory
榜单荣誉：三大记忆力基准 Alfworld、Scienceworld、Embodied bench 取得领先地位
核心突破
AlfWorld
成功率SR(%)EmbodiedBench
成功率SR(%)ScienceWorld
成功率SR(%)评测领先
04 具身操作
Embodied Manipulation
榜单荣誉：操作类顶尖评测集 RoboCasa Composite Tasks 超过强基线
核心突破
RoboCasa Composite Tasks
平均分(avg)25.67z0
vs.4.8official_report
5.3倍
相较于官方基线模型，性能暴涨近 5.3 倍
细分任务表现
z0official_report
基于这一全新范式，z0 在流式视觉理解、具身记忆、组合复杂操作、视觉导航等 8 项行业权威评测中，精度均领先主流方案。在持续交互中，成功率还可提升 20% 以上
连接未来，共同进化
源自清华定义物理智能体未来
最新论文成果
具身多模态感知与理解
- StreamMind: Unlocking Full Frame Rate Streaming Video Dialogue through Event-Gated Cognition (ICCV 2025)
- Babel: A Scalable Pre-trained Model for Multi-Modal Sensing via Expandable Modality Alignment (SenSys 2025)
- Em-Garde: A Propose-Match Framework for Proactive Streaming Video Understanding (arXiv 2026)
- VoLLM: Smoothness-aware Serving of LLM-powered Voice Q&A via Adaptive Preemption (IPDPS 2026)
- Empower Vision Applications with LoRA LMM (VaLoRA) (EuroSys 2025)
- Region-based Content Enhancement for Efficient Video Analytics at the Edge (RegenHance) (NSDI 2025)
- BiSwift: Bandwidth Orchestrator for Multi-Stream Video Analytics on Edge (INFOCOM 2024)
具身记忆与长程推理
- Sculptor: Empowering LLMs with Cognitive Agency via Active Context Management (ICLR 2026)
- SeerAttention: Learning Intrinsic Sparse Attention in Your LLMs (NeurIPS 2025)
- SeerAttention-R: Sparse Attention Adaptation for Long Reasoning (ICLR 2026)
- Time's Up! An Empirical Study of LLM Reasoning Ability Under Output Length Constraint (EMNLP 2025)
- AVA: Towards Agentic Video Analytics with Vision Language Models (NSDI 2026)
- Video-in-the-Loop: Span-Grounded Long Video QA with Interleaved Reasoning (ICML 2026)
- OxyGen: Unified KV Cache Management for Vision-Language-Action Models under Multi-Task Parallelism (arXiv 2026)
- JENGA: Enhancing LLM Long-Context Fine-tuning with Contextual Token Sparsity (USENIX ATC 2025)
具身决策与自主规划
- AdaNav: Adaptive Reasoning with Uncertainty for Vision-Language Navigation (ICLR 2026)
- MetaNav: Stop Wandering: Efficient Vision-Language Navigation via Metacognitive Reasoning (arXiv 2026)
- V-Droid: Advancing Mobile GUI Agents: A Verifier-Driven Approach to Practical Deployment (MobiCom 2026)
- ProRe: A Proactive Reward System for GUI Agents via Reasoner-Actor Collaboration (ICLR 2026)
- A3TP: Automated, Accurate, and Adaptive UAV Task Planning for Large-Scale Power Transmission Networks Inspection (MobiCom 2026)
- Auto-UIT: Automated UAV Inspection Trajectory Generation from Noisy Sparse 3D Point Cloud (MobiCom 2025)
数据合成与质量评估
- Condor: Enhance LLM Alignment with Knowledge-Driven Data Synthesis and Refinement (ACL 2025)
- SynDiSC: High-Quality Tabular Data Synthesis with Distributional and Semantic Consistency (SIGIR 2026)
- Seeing the Whole Through the Parts: Discovering Objects through Semantic Part Mining in Weak Supervision (SIGIR 2026)
- PriCAF: Privacy-Preserving Contribution Assessment in Federated Learning Before Model Training (ACM MM 2025)
- Training Data Attribution: Was Your Model Secretly Trained On Data Created By Mine? (KDD 2025)
- Multi-Label and Evolvable Dataset Preparation for Web-Based Object Detection (TKDD 2024)
- CoAst: Validation-Free Contribution Assessment for Federated Learning based on Cross-Round Valuation (ACM MM 2024)
- VIDAR: Data Quality Improvement for Monocular 3D Reconstruction through In-situ Visual Interaction (ICRA 2024)
高效推理与边缘部署
- Scaling LLM Test-Time Compute with Mobile NPU on Smartphones (EuroSys 2026)
- T-MAC: CPU Renaissance via Table Lookup for Low-Bit LLM Deployment on Edge (EuroSys 2025)
- BitDistiller: Unleashing the Potential of Sub-4-Bit LLMs via Self-Distillation (ACL 2024)
- BitDecoding: Unlocking Tensor Cores for Long-Context LLMs Decoding with Low-Bit KV Cache (HPCA 2026)
- Pre-gated MoE: An Algorithm-System Co-Design for Fast and Scalable Mixture-of-Expert Inference (ISCA 2024)
- Neuralink: Fast on-Device LLM Inference with Neuron Co-Activation Linking (ASPLOS 2025)
- H2O: Heterogeneity-Aware Hierarchical Orchestration for Memory-Efficient On-Device LLM Inference (TMC 2025)
- SwapMoE: Serving Off-the-shelf MoE-based Large Language Models with Tunable Memory Budget (ACL 2024)
- Empowering In-Browser Deep Learning Inference on Edge Through Just-In-Time Kernel Optimization (MobiSys 2024)
- FlexNN: Efficient and Adaptive DNN Inference on Memory-Constrained Edge Devices (MobiCom 2024)
- Amanda: Unified Instrumentation Framework for Deep Neural Networks (ASPLOS 2024)
- LUT Tensor Core: A Software-Hardware Co-Design for LUT-Based Low-Bit LLM Inference (ISCA 2025)
存算一体与新型计算范式
- Pim-dl: Expanding the Applicability of Commodity DRAM-PIMs for Deep Learning via Algorithm-System Co-Optimization (ASPLOS 2024)
- VSPIM: SRAM Processing-in-Memory DNN Acceleration via Vector-Scalar Operations (IEEE ToC 2023)
- Bulk Bitwise Accumulation in Commercial DRAM (NeurIPS 2024 Workshop MLNCP Oral)
- PUDTune: Multi-Level Charging for High-Precision Calibration in Processing-Using-DRAM (IEEE CAL 2025)
- MVDRAM: Enabling GeMV Execution in Unmodified DRAM for Low-Bit LLM Acceleration (arXiv 2025)