# VLA 论文索引 (Paper Index)

> **快速查找**: 多维度索引系统，帮助快速定位相关论文
> **最后更新**: 2026-02-11

---

## 📋 使用指南

> **Industry Note**: Jim Fan 2025 年度复盘（硬件可靠性 / 评测可复现 / VLM→VLA 路线反思）见 [`./frontier/jim_fan_2025_robotics_lessons.md`](../frontier/jim_fan_2025_robotics_lessons.md)
> **Industry Note**: NVIDIA AI 五层蛋糕（Energy / Chips / Infrastructure / Models / Applications）见 [`./frontier/nvidia_ai_5_layer_cake_infrastructure_2026.md`](../deployment/nvidia_ai_5_layer_cake_infrastructure_2026.md)

| 查找方式 | 跳转链接 |
|:---|:---|
| 按技术分类 | [技术分类索引](#技术分类索引) |
| 按公司/机构 | [公司分类索引](#公司分类索引) |
| 按时间线 | [时间线索引](#时间线索引) |
| 完整综述 | [literature_review.md](./literature_review.md) |

---

## 🎯 快速索引表

### 按技术分类索引

| 技术类别 | 论文 | 链接 |
|:---|:---|:---|
| **Diffusion** | Diffusion Policy | [详细](#diffusion-policy) |
| | RDT-1B | [详细](#rdt) |
| | RDT2 | [深度解读](rdt2_umi_zero_shot_cross_embodiment_2026.md) |
| | DKT Perception | [深度解析](../perception/dkt_transparency_perception.md) |
| **Tactile / Visuotactile** 🆕 | VT Pretraining + Online Multitask (SciRobotics 2026) | [深度解析](../tactile/visual_tactile_pretraining_online_multitask_learning_2026.md) |
|  | GenForce (Nat Commun 2026) | [深度笔记](../tactile/genforce_tactile_force_transfer_2026.md) |
|  | TaF-VLA (Tactile-Force Alignment) | [深度解读](../tactile/taf_vla_tactile_force_alignment_2026.md) |
|  | TacRefineNet (Tactile-Only Grasp Refinement) | [深度笔记](../tactile/tacrefinenet_tactile_only_grasp_refinement_2026.md) |
|  | TouchGuide (Inference-Time Steering) | [深度解读](../tactile/touchguide_inference_time_steering_touch_guidance_2026.md) |
| **Physics-Inspired Vision** 🆕 | WaveFormer (Wave Equation) | [深度解读](../perception/waveformer_wave_equation_vision_2026.md) |
| **3D Reconstruction / View Synthesis** 🆕 | Zero-1-to-3 | [深度解读](../perception/zero_1_to_3_zero_shot_one_image_to_3d_object_2023.md) |
| **Evaluation / World Model** 🆕 | WorldEval | [深度解读](../world-model/worldeval_world_model_policy_evaluator_2025.md) |
|  | ENACT | [深度解读](../planning/enact_embodied_cognition_world_modeling_egocentric_interaction_2025.md) |
|  | DreamZero (WAM) | [深度解析](../world-model/dreamzero_world_action_models_zero_shot_policies_2026.md) |
|  | AtomVLA (Offline Post-training) | [深度解读](../world-model/atomvla_offline_post_training_predictive_latent_world_models_2026.md) |
| **Embodied Task Planning / Video Reasoning** 🆕 | Thinker | [深度解读](../planning/thinker_vlm_embodied_intelligence_2026.md) |
|  | RynnBrain | [深度笔记](../vla-core/rynnbrain_open_embodied_foundation_models_2026.md) |
|  | VITRA | [深度解读](../vla-core/vitra_scalable_vla_pretraining_human_activity_videos_2026.md) |
| **Generative Science / Biomolecular Structure** 🆕 | IntelliFold 2 | [深度解读](../frontier/intellifold_2_surpassing_alphafold3_structural_consistency_2026.md) |
| **Quantization / Compression** 🆕 | QVLA | [深度解读](qvla_action_centric_quantization_2026.md) |
| **Graph ML / GNN 基线** 🆕 | Classic GNNs are Strong Baselines | [深度解读](../frontier/classic_gnns_strong_baselines_node_classification_2024.md) |
| **Multimodal Fusion** 🆕 | Policy Consensus (Multi-Modal Manipulation) | [深度解读](../tactile/policy_consensus_multimodal_manipulation_2025.md) |
| **Embodied Data / Policy Iteration** 🆕 | RoboPocket | [深度解读](../deployment/robopocket_robot_free_instant_policy_iteration_phone_2026.md) |
| **Flow Matching** | π0 | [深度解析](../diffusion-flow/pi0_flow_matching.md) |
| | π0.5 | [深度解析](../vla-core/pi0_5_dissection.md) |
| | π0.6 | [深度解析](../vla-core/pi0_6_dissection.md) |
| | LingBot-VLA | [深度解析](../vla-core/lingbot_vla_pragmatic_vla_foundation_model_2026.md) |
| | Shallow-π | [深度解读](shallow_pi_knowledge_distillation_flow_vla_2026.md) |
| **Tokenization** | RT-2 | [深度解析](#rt-2) |
| | OpenVLA | [深度解析](#openvla) |
| | FAST | [详细](#fast) |
| **RL 训练** | GR-RL | [深度解析](../rl/gr_rl_dissection.md) |
| | π*0.6 Recap | [深度解析](../vla-core/pi0_6_dissection.md#recap) |
| | RLinf（VLA+RL Infra） | [工具解读](../rl/rlinf_vla_rl_training.md) |
| **Online RL / Replay** 🆕 | U2O RL (Unsupervised-to-Online) | [深度解读](../rl/unsupervised_to_online_reinforcement_learning_u2o_2024.md) |
|  | PGR (Prioritized Generative Replay) | [深度解读](../rl/prioritized_generative_replay_pgr_2025.md) |
| **架构创新** | WALL-OSS | [深度解析](../vla-core/wall_oss.md) |
| | Galaxea G0 | [详细](#galaxea-g0) |
| **训练技术** | Knowledge Insulation | [摘要](#knowledge-insulation) |
| **Latent Action** | UniVLA | [详细](#univla) |
| | EvoVLA | [详细](#evovla) |
| | MemoryVLA | [详细](#memoryvla) |
| **VLM × 推理/表征** 🆕 | Embodied CoT (ECoT) | [深度解读](../planning/embodied_chain_of_thought_robotic_control_2024.md) |
|  | PR2L (Promptable Representations) | [深度解读](../planning/vlm_promptable_representations_for_rl_pr2l_2025.md) |
| **Reasoning / Test-time Scaling** 🆕 | DAC-RL (Divide-and-Conquer) | [深度解读](../planning/dac_rl_divide_conquer_reasoning_2026.md) |
|  | TinyLoRA (13 params) | [深度解读](../planning/tiny_lora_13_params_reasoning_2026.md) |
| **NeurIPS 2025** 🆕 | Artificial Hivemind | [详细](./neurips_2025_insights.md#1-artificial-hivemind-语言模型的同质化问题) |
| | Gated Attention | [详细](./neurips_2025_insights.md#2-gated-attention-门控注意力机制) |
| | 1000 Layer Networks | [详细](./neurips_2025_insights.md#3-1000-layer-networks-深层自监督-rl) |
| | Diffusion Generalization | [详细](./neurips_2025_insights.md#4-diffusion-models-的泛化机制) |
| | Superposition Scaling | [详细](./neurips_2025_insights.md#5-superposition-表示叠加与神经缩放) |
| | RL Reasoning Limits | [详细](./neurips_2025_insights.md#6-rlvr-的局限性rl-真的能扩展推理能力吗) |

### 按公司分类索引

| 公司/机构 | 论文 | 链接 |
|:---|:---|:---|
| **Google DeepMind** | RT-2 | [深度解析](#rt-2) |
| | RT-1 | [详细](#rt-1) |
| **Physical Intelligence** | π0 | [深度解析](../diffusion-flow/pi0_flow_matching.md) |
| | π0.5 | [深度解析](../vla-core/pi0_5_dissection.md) |
| | π0.6 | [深度解析](../vla-core/pi0_6_dissection.md) |
| | FAST | [详细](#fast) |
| | Knowledge Insulation | [摘要](#knowledge-insulation) |
| **ByteDance Seed** | GR-RL | [深度解析](../rl/gr_rl_dissection.md) |
| | RDT-1B | [详细](#rdt) |
| **NVIDIA** 🆕 | DreamZero (WAM) | [深度解析](../world-model/dreamzero_world_action_models_zero_shot_policies_2026.md) |
| **RDT Team / 清华大学 MARS Lab** 🆕 | RDT2 | [深度解读](rdt2_umi_zero_shot_cross_embodiment_2026.md) |
| **Stanford** | OpenVLA | [深度解析](#openvla) |
| | ACT | [详细](#act) |
| **X² (自变量)** | WALL-OSS | [深度解析](../vla-core/wall_oss.md) |
| **Galaxea AI** | G0 | [详细](#galaxea-g0) |
| **Robbyant Team** 🆕 | LingBot-VLA | [深度解析](../vla-core/lingbot_vla_pragmatic_vla_foundation_model_2026.md) |
| **优必选（UBTECH Robotics）** 🆕 | Thinker | [深度解读](../planning/thinker_vlm_embodied_intelligence_2026.md) |
| **阿里达摩院（DAMO Academy）** 🆕 | RynnBrain | [深度笔记](../vla-core/rynnbrain_open_embodied_foundation_models_2026.md) |
| **Xiaomi Robotics** 🆕 | TacRefineNet | [深度笔记](../tactile/tacrefinenet_tactile_only_grasp_refinement_2026.md) |
| **IntelliGen-AI** 🆕 | IntelliFold 2 | [深度解读](../frontier/intellifold_2_surpassing_alphafold3_structural_consistency_2026.md) |
| **Samsung Research** 🆕 | Shallow-π | [深度解读](shallow_pi_knowledge_distillation_flow_vla_2026.md) |
| **UCLA / Microsoft** 🆕 | DAC-RL | [深度解读](../planning/dac_rl_divide_conquer_reasoning_2026.md) |
| **Microsoft Research Asia** 🆕 | VITRA | [深度解读](../vla-core/vitra_scalable_vla_pretraining_human_activity_videos_2026.md) |
| **北航 / 香港理工大学** 🆕 | Classic GNNs are Strong Baselines | [深度解读](../frontier/classic_gnns_strong_baselines_node_classification_2024.md) |
| **Columbia** | Diffusion Policy | [详细](#diffusion-policy) |
| | Zero-1-to-3 | [深度解读](../perception/zero_1_to_3_zero_shot_one_image_to_3d_object_2023.md) |
| **WorldEval Team** 🆕 | WorldEval | [深度解读](../world-model/worldeval_world_model_policy_evaluator_2025.md) |
| **SJTU AutoLab / Anyverse / CAS / Ant Group** 🆕 | QVLA | [深度解读](qvla_action_centric_quantization_2026.md) |
| **Meta FAIR** 🆕 | TinyLoRA | [深度解读](../planning/tiny_lora_13_params_reasoning_2026.md) |
| **UIUC / Harvard / MIT / Columbia** 🆕 | WaveFormer | [深度解读](../perception/waveformer_wave_equation_vision_2026.md) |
| **UIUC / Harvard / MIT / Columbia** 🆕 | Policy Consensus | [深度解读](../tactile/policy_consensus_multimodal_manipulation_2025.md) |
| **上海交大 / Xense Robotics / 中山大学 / Oxford / UCSD** 🆕 | TouchGuide | [深度解读](../tactile/touchguide_inference_time_steering_touch_guidance_2026.md) |
| **浙江大学** 🆕 | DKT Perception | [深度解析](../perception/dkt_transparency_perception.md) |
| | VT Pretraining + Online Multitask (SciRobotics 2026) | [深度解析](../tactile/visual_tactile_pretraining_online_multitask_learning_2026.md) |
| **华盛顿大学** 🆕 | Artificial Hivemind | [详细](./neurips_2025_insights.md#1-artificial-hivemind-语言模型的同质化问题) |
| **阿里千问** 🆕 | Gated Attention | [详细](./neurips_2025_insights.md#2-gated-attention-门控注意力机制) |
| **普林斯顿** 🆕 | 1000 Layer Networks | [详细](./neurips_2025_insights.md#3-1000-layer-networks-深层自监督-rl) |
| **巴黎 PSL** 🆕 | Diffusion Generalization | [详细](./neurips_2025_insights.md#4-diffusion-models-的泛化机制) |
| **MIT** 🆕 | Superposition Scaling | [详细](./neurips_2025_insights.md#5-superposition-表示叠加与神经缩放) |
| **清华大学** 🆕 | RL Reasoning Limits | [详细](./neurips_2025_insights.md#6-rlvr-的局限性rl-真的能扩展推理能力吗) |

### 按时间线索引

| 年份 | 论文 | 链接 |
|:---|:---|:---|
| **2023** | Diffusion Policy | [详细](#diffusion-policy) |
| | RT-2 | [深度解析](#rt-2) |
| | ACT | [详细](#act) |
| | Zero-1-to-3 | [深度解读](../perception/zero_1_to_3_zero_shot_one_image_to_3d_object_2023.md) |
| **2024** | OpenVLA | [深度解析](#openvla) |
| | π0 | [深度解析](../diffusion-flow/pi0_flow_matching.md) |
| | RDT-1B | [详细](#rdt) |
| | DKT Perception | [深度解析](../perception/dkt_transparency_perception.md) |
| | Galaxea G0 | [详细](#galaxea-g0) |
| | Knowledge Insulation | [摘要](#knowledge-insulation) |
| | Classic GNNs are Strong Baselines | [深度解读](../frontier/classic_gnns_strong_baselines_node_classification_2024.md) |
| | UniVLA | [详细](#univla) |
| | Embodied CoT (ECoT) | [深度解读](../planning/embodied_chain_of_thought_robotic_control_2024.md) |
| | PR2L (Promptable Reps for RL) | [深度解读](../planning/vlm_promptable_representations_for_rl_pr2l_2025.md) |
| | U2O RL (Unsupervised-to-Online) | [深度解读](../rl/unsupervised_to_online_reinforcement_learning_u2o_2024.md) |
| **2025** | π0.5 | [深度解析](../vla-core/pi0_5_dissection.md) |
| | VITRA | [深度解读](../vla-core/vitra_scalable_vla_pretraining_human_activity_videos_2026.md) |
| | π0.6 | [深度解析](../vla-core/pi0_6_dissection.md) |
| | FAST | [详细](#fast) |
| | GR-RL | [深度解析](../rl/gr_rl_dissection.md) |
| | WALL-OSS | [深度解析](../vla-core/wall_oss.md) |
| | EvoVLA | [详细](#evovla) |
| | MemoryVLA | [详细](#memoryvla) |
| | TTF-VLA | [详细](#ttf-vla) |
| | OmniVLA | [详细](#omnivla) |
| | MergeVLA | [详细](#mergevla) |
| | PGR (Prioritized Generative Replay) | [深度解读](../rl/prioritized_generative_replay_pgr_2025.md) |
| | Policy Consensus (Multi-Modal Manipulation) | [深度解读](../tactile/policy_consensus_multimodal_manipulation_2025.md) |
| | WorldEval | [深度解读](../world-model/worldeval_world_model_policy_evaluator_2025.md) |
| | **NeurIPS 2025 Best Papers** 🆕 | [专题解读](./neurips_2025_insights.md) |
| **2026** 🆕 | LingBot-VLA | [深度解析](../vla-core/lingbot_vla_pragmatic_vla_foundation_model_2026.md) |
| | DreamZero (WAM) | [深度解析](../world-model/dreamzero_world_action_models_zero_shot_policies_2026.md) |
| | RDT2 (UMI Zero-shot) | [深度解读](rdt2_umi_zero_shot_cross_embodiment_2026.md) |
| | Shallow-π (Distillation) | [深度解读](shallow_pi_knowledge_distillation_flow_vla_2026.md) |
| | DAC-RL (Divide-and-Conquer) | [深度解读](../planning/dac_rl_divide_conquer_reasoning_2026.md) |
| | TinyLoRA (13 params) | [深度解读](../planning/tiny_lora_13_params_reasoning_2026.md) |
| | Thinker (Embodied Planning VLM) | [深度解读](../planning/thinker_vlm_embodied_intelligence_2026.md) |
| | RynnBrain (Embodied Foundation Model) | [深度笔记](../vla-core/rynnbrain_open_embodied_foundation_models_2026.md) |
| | TacRefineNet (Tactile-Only Grasp Refinement) | [深度笔记](../tactile/tacrefinenet_tactile_only_grasp_refinement_2026.md) |
| | AtomVLA (Offline Post-training) | [深度解读](../world-model/atomvla_offline_post_training_predictive_latent_world_models_2026.md) |
| | RoboPocket (Instant Policy Iteration) | [深度解读](../deployment/robopocket_robot_free_instant_policy_iteration_phone_2026.md) |
| | IntelliFold 2 (Generative Science) | [深度解读](../frontier/intellifold_2_surpassing_alphafold3_structural_consistency_2026.md) |
| | QVLA (Quantization) | [深度解读](qvla_action_centric_quantization_2026.md) |
| | TaF-VLA (Tactile-Force Alignment) | [深度解读](../tactile/taf_vla_tactile_force_alignment_2026.md) |
| | TouchGuide (Inference-Time Steering) | [深度解读](../tactile/touchguide_inference_time_steering_touch_guidance_2026.md) |
| | VT Pretraining + Online Multitask (SciRobotics 2026) | [深度解析](../tactile/visual_tactile_pretraining_online_multitask_learning_2026.md) |
| | Video Generation Models in Robotics (Survey) | [前沿笔记](../world-model/video_generation_models_in_robotics_survey_2026.md) |
| | WaveFormer (Wave Equation) | [深度解读](../perception/waveformer_wave_equation_vision_2026.md) |

---

## 🔍 详细分类索引

### 技术分类索引

#### 0. 图学习基线与评测 (Graph ML Baselines)

- **Classic GNNs are Strong Baselines** (NeurIPS 2024)
  - 主题: 经典 GNN 基线再评估、超参敏感性、评测公平性
  - [深度解读](../frontier/classic_gnns_strong_baselines_node_classification_2024.md)

#### 0.1 3D 重建与新视角合成 (3D Reconstruction / View Synthesis)

- **Zero-1-to-3** (arXiv 2023)
  - 主题: 单图像新视角合成 + 3D 重建先验
  - [深度解读](../perception/zero_1_to_3_zero_shot_one_image_to_3d_object_2023.md)

#### 0.2 评测与世界模型 (Evaluation / World Model)

- **WorldEval** (arXiv 2025)
  - 主题: 世界模型评估机器人策略、生成式 rollout
  - [深度解读](../world-model/worldeval_world_model_policy_evaluator_2025.md)

- **ENACT** (arXiv 2025 / ICLR 2026)
  - 主题: 用 egocentric interaction world modeling 测 VLM 是否具备具身认知、action-effect reasoning 与长时程 memory
  - [深度解读](../planning/enact_embodied_cognition_world_modeling_egocentric_interaction_2025.md)

- **2025 BEHAVIOR Challenge 冠军方案** (arXiv 2025)
  - 主题: benchmark 驱动的 system recipe；在 Pi0.5 上加入 correlated noise、System 2 stage tracking、action compression 与 correction rules
  - [深度解读](../planning/behavior_challenge_2025_first_place_solution_task_adaptation_vla_2025.md)

- **IS-Bench** (arXiv 2025 / AAAI 2026)
  - 主题: 交互式 embodied safety benchmark；测风险识别与 risk mitigation 是否在正确时机发生，而不是只看终局是否安全
  - [深度解读](../planning/is_bench_interactive_safety_vlm_embodied_agents_household_tasks_2025.md)

- **DreamZero** (2026)
  - 主题: World Action Model (WAM)，联合预测视频 + 动作，基于 video diffusion 的零样本策略
  - [深度解析](../world-model/dreamzero_world_action_models_zero_shot_policies_2026.md)

- **AtomVLA** (arXiv 2026)
  - 主题: 原子子任务分解 + predictive latent world model + offline GRPO；把 VLA 后训练从真机试错推进到离线 latent evaluator
  - [深度解读](../world-model/atomvla_offline_post_training_predictive_latent_world_models_2026.md)

#### 0.25 数据采集与即时策略迭代 (Data Collection / Instant Policy Iteration)

- **RoboPocket** (arXiv 2026)
  - 主题: 用手机 AR 直接可视化策略意图，在没有真实机器人在场的情况下做 DAgger-like corrective data collection 与分钟级在线微调
  - [深度解读](../deployment/robopocket_robot_free_instant_policy_iteration_phone_2026.md)

#### 0.3 量化与模型压缩 (Quantization / Compression)

- **QVLA** (arXiv 2026)
  - 主题: 动作空间敏感量化、通道级比特分配与剪枝一体化
  - [深度解读](qvla_action_centric_quantization_2026.md)

#### 0.4 具身任务规划与 Ego-view 视频理解 (Embodied Task Planning / Ego-view Video)

- **RynnBrain** (Release 2026)
  - 主题: 物理现实锚定的具身基础模型（定位/指向/轨迹/规划统一输出）+ RynnBrain-Bench（Object/Spatial/Grounding/Pointing）
  - [深度笔记](../vla-core/rynnbrain_open_embodied_foundation_models_2026.md)

- **Thinker** (arXiv 2026)
  - 主题: 具身 VLM 任务规划、第一视角空间理解、视频末帧关键帧增强（keyframe + video）
  - [深度解读](../planning/thinker_vlm_embodied_intelligence_2026.md)

- **VITRA** (arXiv 2025)
  - 主题: 无标注人类手部 egocentric 视频 → VLA 同构 episode（指令 + 3D 手部运动 + 相机运动），可扩展预训练
  - [深度解读](../vla-core/vitra_scalable_vla_pretraining_human_activity_videos_2026.md)

#### 0.5 生成式科学智能与生物结构预测 (Generative Science / Biomolecular Structure)

- **IntelliFold 2** (Release Note 2026)
  - 主题: Ab-Ag / Protein-Ligand 共折叠；latent space scaling、随机原子化（stochastic atomization）、（Pro）PPO 稳采样与难例损失加权
  - [深度解读](../frontier/intellifold_2_surpassing_alphafold3_structural_consistency_2026.md)

#### 1. 动作生成策略

##### Diffusion 系列
- **Diffusion Policy** (2023)
  - 技术: DDPM, U-Net/DiT
  - 动作空间: 连续
  - [详细内容](./literature_review.md#1-diffusion-policy)

- **RDT-1B** (2024)
  - 技术: DiT, 十亿参数
  - 动作空间: 连续
  - [详细内容](../vla-core/rdt.md)

- **RDT2** (2026)
  - 技术: Residual VQ + Flow Matching + 单步蒸馏
  - 动作空间: 连续
  - [深度解读](rdt2_umi_zero_shot_cross_embodiment_2026.md)

##### Flow Matching 系列
- **π0** (2024)
  - 技术: Flow Matching, ODE Solver
  - 动作空间: 连续
  - [深度解析](../diffusion-flow/pi0_flow_matching.md) | [代码解析](../vla-core/pi0_code_analysis.md)

- **π0.5** (2025)
  - 技术: Flow Matching + 隐式推理
  - 动作空间: 连续
  - [深度解析](../vla-core/pi0_5_dissection.md)

- **π0.6** (2025)
  - 技术: Flow Matching + Action Expert
  - 动作空间: 连续
  - [深度解析](../vla-core/pi0_6_dissection.md)

- **LingBot-VLA** (2026)
  - 技术: Flow Matching + VLM + Action Expert（共享 attention）；可选 depth distill
  - 动作空间: 连续
  - [深度解析](../vla-core/lingbot_vla_pragmatic_vla_foundation_model_2026.md)

##### Tokenization 系列
- **RT-2** (2023)
  - 技术: Action Tokenization, 256-bins
  - 动作空间: 离散
  - [详细内容](./literature_review.md#2-rt-2)

- **OpenVLA** (2024)
  - 技术: Action Tokenization + LoRA
  - 动作空间: 离散
  - [详细内容](./literature_review.md#3-openvla)

- **FAST** (2025)
  - 技术: DCT + BPE Tokenization
  - 动作空间: 离散（压缩）
  - [详细内容](../vla-core/fast.md)

##### 其他
- **ACT** (2023)
  - 技术: CVAE + Action Chunking
  - 动作空间: 连续
  - [详细内容](../vla-core/act.md)

##### Latent Action 系列 🆕
- **UniVLA** (IJRR 2024)
  - 技术: Task-centric Latent Actions
  - 核心: 从视频学习机器人无关的动作表示
  - [详细内容](./literature_review.md#univla-ijrr-2024)

- **EvoVLA** (2025)
  - 技术: Self-Evolving + Stage-Aligned Reward
  - 核心: 解决长时程"阶段幻觉"问题
  - [详细内容](./literature_review.md#evovla-2025)

- **MemoryVLA** (2025)
  - 技术: Perception-Cognition Memory
  - 核心: 感知-认知记忆系统
  - [详细内容](./literature_review.md#memoryvla-2025)

- **TTF-VLA** (2025)
  - 技术: Temporal Token Fusion
  - 核心: 训练无关的多帧融合

- **OmniVLA** (2025)
  - 技术: Multi-sensor Perception
  - 核心: 红外/雷达/麦克风多传感器融合

- **MergeVLA** (2025)
  - 技术: Cross-skill Model Merging
  - 核心: 跨技能知识迁移

#### 2. 训练方法

##### BC (Behavior Cloning)
- **RT-2** - Co-fine-tuning
- **OpenVLA** - LoRA Fine-tuning
- **π0** - Flow Matching Training

##### RL (Reinforcement Learning)
- **GR-RL** (2025)
  - 技术: Offline + Online RL
  - [深度解析](../rl/gr_rl_dissection.md)

- **π*0.6 Recap** (2025)
  - 技术: Offline RL
  - [深度解析](../vla-core/pi0_6_dissection.md#recap)

- **RLinf** (2025)
  - 定位: Reinforcement Learning Infrastructure（面向 Embodied / Agentic AI）
  - 关联: VLA+RL 训练框架（RLinf-VLA）、真机 RL、分布式 rollout 与可复现评估
  - [项目主页](https://github.com/RLinf/RLinf) | [手册解读](../rl/rlinf_vla_rl_training.md)

##### 轨迹优化 / Guided Policy Search（经典）
> 注：这些是 VLA 之前的“真机学习底座”，但它们解释了为什么高精度任务往往需要“可控老师 + 可泛化学生 + 闭环”。

- **Optimal Feedback Control as a Theory of Motor Coordination** (Todorov & Jordan, Nat Neurosci 2002)
  - 技术: stochastic optimal feedback control + minimal intervention principle + task-relevant correction
  - 手册解读（笔记已下线） | [论文 DOI](https://doi.org/10.1038/nn963)

- **Guided Policy Search under Unknown Dynamics** (Levine & Abbeel, NeurIPS 2014)
  - 技术: local linear dynamics + KL trust region + distillation
  - 手册解读（笔记已下线） | [论文 PDF](https://proceedings.neurips.cc/paper_files/paper/2014/file/6766aa2750c19aad2fa1b32f36ed4aee-Paper.pdf)

- **End-to-End Training of Deep Visuomotor Policies** (Levine et al., JMLR 2016)
  - 技术: GPS 把 RL 转写成监督学习，学习像素→扭矩策略
  - 手册解读（笔记已下线） | [arXiv](https://arxiv.org/abs/1504.00702)

##### 混合方法
- **π0.5** - Co-training (Robot + Internet + Sim)

#### 3. 架构创新

##### 单模型架构
- **RT-2** - VLM + Action Tokens
- **OpenVLA** - VLM + Action Head
- **π0** - VLM + Flow Matching

##### 双系统架构
- **Galaxea G0** (2024)
  - G0-VLM (大脑) + G0-VLA (小脑)
  - [详细内容](../vla-core/galaxea_g0.md)

- **π0.6** - VLM (大脑) + Action Expert (小脑)
  - [深度解析](../vla-core/pi0_6_dissection.md#action-expert)

##### 层级架构
- **WALL-OSS** (2025) - A级
  - Hierarchical CoT + Dual Heads
  - [详细内容](../vla-core/wall_oss.md)

#### 4. 应用场景

##### 操作任务
- RT-2, OpenVLA, π0, GR-RL

##### 导航任务
- (待补充)

##### 灵巧手
- GR-RL (穿鞋带), RDT-1B (双臂操作)
- UniTacHand (2025): MANO UV Map 统一触觉表征，实现人手→机器人灵巧手策略迁移（零样本/小样本）
  - [手册解读](../tactile/unitachhand.md)
  - [论文（HTML）](https://arxiv.org/html/2512.21233v2)

##### 触觉 / 视触觉 (Tactile)
- 触觉为什么不可替代（行业/研究盘点）：[笔记](../tactile/tactile_irreplaceable.md)
- **Grip Force Classic** (Brain 2004): 在长期去传入感觉条件下，抓取力会退化为高力、滞后、低精度兜底；说明 predictive grip-force control 需要至少间歇性的 somatosensory feedback 来校准 internal model
  - 经典笔记（笔记已下线）
  - [论文 DOI](https://doi.org/10.1093/brain/awh016)
- **TacRefineNet** (arXiv 2025 / 索引归入 2026 前沿): 多指触觉 + 本体融合的目标驱动微调，迭代回归 wrist 6DoF 增量，面向抓取执行“最后一公里”误差补偿
  - [深度笔记](../tactile/tacrefinenet_tactile_only_grasp_refinement_2026.md)
  - [论文 PDF](https://arxiv.org/pdf/2509.25746)
  - [Project Page](https://sites.google.com/view/tacrefinenet)
- **TouchGuide** (2026): 在动作采样末段用 Contact Physical Model 做触觉引导，不重训 base policy；配套 TacUMI 解决高质量 contact-rich 示教采集
  - [深度解读](../tactile/touchguide_inference_time_steering_touch_guidance_2026.md)
  - [论文 HTML](https://arxiv.org/html/2601.20239v3)
- **SuperTac + DOVE** (Nature Sensors, 2025): 多模态电子皮肤（多光谱+摩擦电+IMU+温度/接近/振动）+ 触觉语言模型
  - [论文页](https://www.nature.com/articles/s44460-025-00006-y)
  - [手册笔记](../../deployment/perception/supertac_dove_biomimetic_multimodal_tactile_sensing.md)
- **VLA-Touch** (2025): Dual-Level Tactile Feedback
  - [论文](https://arxiv.org/abs/2507.17294)
- **OmniVTLA** (2025): Semantic-Aligned Tactile Sensing
  - [论文](https://arxiv.org/abs/2508.08706)
- **GelSLAM** (2025): Real-time High-Fidelity 3D Tactile SLAM
  - [论文](https://arxiv.org/abs/2508.15990)
- **Tactile Robotics: An Outlook** (2025): 触觉机器人综述与路线图
  - [论文](https://arxiv.org/abs/2508.11261)

---

### 公司分类索引

#### Google DeepMind
- **RT-2** (2023)
- **RT-1** (2022)

#### Physical Intelligence
- **π0** (2024)
- **π0.5** (2025)
- **π0.6** (2025)
- **FAST** (2025)
- **Knowledge Insulation** (2024)

#### ByteDance Seed
- **GR-RL** (2025)
- **RDT-1B** (2024) (与清华合作)

#### RDT Team / 清华大学 MARS Lab
- **RDT2** (2026)

#### Stanford
- **OpenVLA** (2024)
- **ACT** (2023)

#### X² (自变量)
- **WALL-OSS** (2025)

#### Galaxea AI
- **G0** (2024)

#### 阿里达摩院（Alibaba DAMO Academy） 🆕
- **RynnBrain** (Release 2026)
  - [深度笔记](../vla-core/rynnbrain_open_embodied_foundation_models_2026.md)

#### Xiaomi Robotics 🆕
- **TacRefineNet** (arXiv 2025 / 索引归入 2026 前沿)
  - [深度笔记](../tactile/tacrefinenet_tactile_only_grasp_refinement_2026.md)

#### WorldEval Team
- **WorldEval** (2025)
  - [深度解读](../world-model/worldeval_world_model_policy_evaluator_2025.md)

#### SJTU AutoLab / Anyverse / CAS / Ant Group
- **QVLA** (2026)
  - [深度解读](qvla_action_centric_quantization_2026.md)

#### 上海交大 / Xense Robotics / 中山大学 / Oxford / UCSD
- **TouchGuide** (2026)
  - [深度解读](../tactile/touchguide_inference_time_steering_touch_guidance_2026.md)

#### 原力无限
- **AtomVLA** (2026)
  - [深度解读](../world-model/atomvla_offline_post_training_predictive_latent_world_models_2026.md)

#### 上海交大 MVIG / 上海创智学院 / Noematrix
- **RoboPocket** (2026)
  - [深度解读](../deployment/robopocket_robot_free_instant_policy_iteration_phone_2026.md)

#### Columbia / Toyota Research Institute
- **Zero-1-to-3** (2023)
  - [深度解读](../perception/zero_1_to_3_zero_shot_one_image_to_3d_object_2023.md)

#### 北航 / 香港理工大学
- **Classic GNNs are Strong Baselines** (2024)
  - [深度解读](../frontier/classic_gnns_strong_baselines_node_classification_2024.md)

---

### 时间线索引

#### 2023（早期探索）
- Diffusion Policy (RSS 2023)
- RT-2 (ICRA 2023)
- ACT (2023)
- Zero-1-to-3 (arXiv 2023)
  - [笔记](../perception/zero_1_to_3_zero_shot_one_image_to_3d_object_2023.md)

#### 2024（爆发期）
- OpenVLA (2024.06)
- π0 (2024.10)
- RDT-1B (2024.10)
- Galaxea G0 (2024.09)
- Knowledge Insulation (2024)
- Classic GNNs are Strong Baselines (NeurIPS 2024)
  - [笔记](../frontier/classic_gnns_strong_baselines_node_classification_2024.md)

#### 2025（最新进展）
- π0.5 (2025.01)
- π0.6 (2025.11)
- FAST (2025.01)
- GR-RL (2025)
- WALL-OSS (2025)
- WorldEval (2025): 世界模型评测器
  - [笔记](../world-model/worldeval_world_model_policy_evaluator_2025.md)

#### 2026（世界模型 / 视频生成综述）
- RynnBrain (Release 2026): 物理现实锚定的具身基础模型（定位/指向/轨迹/规划）+ RynnBrain-Bench  
  - [笔记](../vla-core/rynnbrain_open_embodied_foundation_models_2026.md)
- TacRefineNet (arXiv 2025 / 索引归入 2026): 仅触觉的抓取执行末端微调框架，目标驱动回归 wrist 6DoF 增量  
  - [笔记](../tactile/tacrefinenet_tactile_only_grasp_refinement_2026.md)
- TouchGuide (arXiv 2026): 推理期触觉引导框架；用 Contact Physical Model 在动作采样末段纠偏，并配套 TacUMI 做高质量视触觉示教采集  
  - [笔记](../tactile/touchguide_inference_time_steering_touch_guidance_2026.md)
- AtomVLA (2026.03): 原子子任务分解 + predictive latent world model 评分 + offline GRPO，把 VLA 后训练从真机试错推进到离线 latent evaluator  
  - [笔记](../world-model/atomvla_offline_post_training_predictive_latent_world_models_2026.md)
- RoboPocket (2026.03): 手机端 AR 可视化策略意图 + 远程推理 + 在线微调，把 DAgger 式纠错推进到 robot-free instant policy iteration  
  - [笔记](../deployment/robopocket_robot_free_instant_policy_iteration_phone_2026.md)
- RDT2 (2026.02): UMI 数据规模化与跨本体零样本部署  
  - [笔记](rdt2_umi_zero_shot_cross_embodiment_2026.md)
- Shallow-π (2026.01): Flow-based VLA 层深蒸馏（18→6）  
  - [笔记](shallow_pi_knowledge_distillation_flow_vla_2026.md)
- DAC-RL (2026.02): 分治推理训练提升测试时可扩展性  
  - [笔记](../planning/dac_rl_divide_conquer_reasoning_2026.md)
- QVLA (2026.02): 动作空间敏感量化与通道级比特分配  
  - [笔记](qvla_action_centric_quantization_2026.md)
- Video Generation Models in Robotics (Survey, arXiv 2026): 视频生成模型在机器人中的应用、挑战与未来方向  
  - [笔记](../world-model/video_generation_models_in_robotics_survey_2026.md)
  - [论文](https://arxiv.org/abs/2601.07823)

---

## 📝 论文添加指南

### 如何添加新论文？

1. **确定文档类型**
   - 深度解析: 开创性、影响大 → 创建独立 `*_dissection.md`
   - 详细章节: 有创新点 → 在 `literature_review.md` 详细章节
   - 简短摘要: 参考价值 → 简短摘要
   - 仅索引: 保持关注 → 仅列标题+链接

2. **分类标记**
   - 技术分类: Diffusion/Flow/Tokenization/RL/架构
   - 公司分类: 按机构归类
   - 时间分类: 按年份

3. **更新索引**
   - 在 `paper_index.md` 相应分类中添加
   - 更新 `literature_review.md`（如需要）
   - 更新 `README.md`（如创建新文档）

### 分类标准

**技术分类**:
- 动作生成: Diffusion, Flow Matching, Tokenization, CVAE, 其他
- 训练方法: BC, RL (Offline/Online), 混合
- 架构: 单模型, 双系统, 层级
- 应用: 操作, 导航, 灵巧手

---

## 🔗 相关资源

- [完整文献综述](./literature_review.md) - 详细技术归纳
- [模型深度解析](./README.md#-part-5-模型详解-model-zoo) - 独立深度解析文档
- [理论总览](./README.md) - 返回理论目录

---

**最后更新**: 2026-02-08
**维护者**: VLA Handbook Team

## 📄 Daily Papers (Auto)

> 本节由日报任务自动追加（只追加，不改动其他内容）。
> 若你要手动维护索引，请在上方既有结构中编辑；此处仅作为每日入口与追踪。


### VLA 架构

| 论文 | 链接 | 备注 |
|:---|:---|:---|
| EverydayVLA: Cost-Effective 6-DOF Manipulator with Unified VLA Model | [link](https://arxiv.org/abs/2511.05397) | 🔧 daily 2026-02-10 \| repo: https://github.com/EverydayRobotics/EverydayVLA |
| TaF-VLA | [link](https://arxiv.org/abs/2601.20321) | 🎯 tactile \| 📖 daily 2026-02-11 |
| TIC-VLA | [link](https://arxiv.org/abs/2602.02459) | 📖 daily 2026-02-11 |
| CAPER: Constrained and Procedural Reasoning for Robotic Scientific Experiments | [link](https://arxiv.org/abs/2602.09367) | 📖 daily 2026-02-12 |
| Sci-VLA: Agentic VLA Inference Plugin for Long-Horizon Tasks in Scientific Experiments | [link](https://arxiv.org/abs/2602.09430) | 📖 daily 2026-02-12 |
| Say, Dream, and Act: Learning Video World Models for Instruction-Driven Robot Manipulation | [link](https://arxiv.org/abs/2602.10717) | 📖 daily 2026-02-13 |
| JEPA-VLA: Video Predictive Embedding is Needed for VLA Models | [link](https://tldr.takara.ai/p/2602.11832) | ⚡ daily 2026-02-14 |
| ABot-N0: Technical Report on the VLA Foundation Model for Versatile Embodied Navigation | [link](https://tldr.takara.ai/p/2602.11598) | ⚡ daily 2026-02-15 |
| UniManip: General-Purpose Zero-Shot Robotic Manipulation with Agentic Operational Graph | [link](https://arxiv.org/abs/2602.13086) | 📖 daily 2026-02-17 |
| Self-Refining Vision Language Model for Robotic Failure Detection and Reasoning | [link](https://arxiv.org/abs/2602.12405) | 📖 daily 2026-02-17 |
| AsyncVLA: An Asynchronous VLA for Fast and Robust Navigation on the Edge | [link](https://arxiv.org/abs/2602.13476) | 📖 daily 2026-02-18 |
| BPP: Long-Context Robot Imitation Learning by Focusing on Key History Frames | [link](https://arxiv.org/abs/2602.15010) | 📖 daily 2026-02-18 |
| Selective Perception for Robot: Task-Aware Attention in Multimodal VLA | [link](https://arxiv.org/abs/2602.15543) | 🔧 daily 2026-02-19 |
| FUTURE-VLA: Forecasting Unified Trajectories Under Real-time Execution | [link](https://arxiv.org/abs/2602.15882) | 📖 daily 2026-02-20 |

### 動作生成

| 论文 | 链接 | 备注 |
|:---|:---|:---|
| StreamVLA: Breaking the Reason-Act Cycle via Completion-State Gating | [link](https://arxiv.org/abs/2602.01100) | 📖 daily 2026-02-10 |
| From Knowing to Doing Precisely: A General Self-Correction and Termination Framework for VLA models | [link](https://arxiv.org/abs/2602.01811) | 📖 daily 2026-02-10 |
| Vlaser: Coupling VLM for Embodied Reasoning with Flow-Matching Action Expert | [link](https://openreview.net/forum?id=8xTDnj39Ti) | 🔧 daily 2026-02-10 \| repo: https://github.com/vlaser-team/vlaser |
| MVISTA-4D: View-Consistent 4D World Model with Test-Time Action Inference for Robotic Manipulation | [link](https://tldr.takara.ai/p/2602.09878) | 📖 daily 2026-02-12 |
| Preference Aligned Visuomotor Diffusion Policies for Deformable Object Manipulation | [link](https://arxiv.org/abs/2602.09583) | 📖 daily 2026-02-12 |
| AutoFly: Vision-Language-Action Model for UAV Autonomous Navigation in the Wild | [link](https://arxiv.org/abs/2602.09657) | 📖 daily 2026-02-12 |
| Rethinking Visual-Language-Action Model Scaling: Alignment, Mixture, and Regularization | [link](https://arxiv.org/abs/2602.09722) | 📖 daily 2026-02-12 |
| RISE: Self-Improving Robot Policy with Compositional World Model | [link](https://arxiv.org/abs/2602.11075) | 🔧 daily 2026-02-13 |
| OAT: Ordered Action Tokenization | [link](https://arxiv.org/abs/2602.04215) | 📖 daily 2026-02-13 |
| Olaf-World: Orienting Latent Actions for Video World Modeling | [link](https://tldr.takara.ai/p/2602.10104) | ⚡ daily 2026-02-15 |
| Flow-Enabled Generalization to Human Demonstrations in Few-Shot Imitation Learning | [link](https://tldr.takara.ai/p/2602.10594) | 🔧 daily 2026-02-15 |
| CausalGDP: Causality-Guided Diffusion Policies for Reinforcement Learning | [link](https://tldr.takara.ai/p/2602.09207) | ⚡ daily 2026-02-16 |
| What Matters in Building Vision-Language-Action Models for Generalist Robots | [link](https://arxiv.org/abs/2412.14058) | 📖 daily 2026-02-17 |
| SteerVLA: Steering Vision-Language-Action Models in Long-Tail Driving Scenarios | [link](https://arxiv.org/abs/2602.08440) | 📖 daily 2026-02-17 |
| Imitating What Works: Simulation-Filtered Modular Policy Learning from Human Videos | [link](https://arxiv.org/abs/2602.13197) | 📖 daily 2026-02-17 |
| DM0: An Embodied-Native Vision-Language-Action Model towards Physical AI | [link](https://arxiv.org/abs/2602.14974) | 📖 daily 2026-02-18 |
| FlowHOI: Flow-based Semantics-Grounded Generation of Hand-Object Interactions for Dexterous Robot Manipulation | [link](https://arxiv.org/abs/2602.13444) | 📖 daily 2026-02-18 |
| ActionCodec: What Makes for Good Action Tokenizers | [link](https://arxiv.org/abs/2602.15397) | ⚡ daily 2026-02-19 |
| World Action Models are Zero-shot Policies | [link](https://arxiv.org/abs/2602.15922) | ⚡ daily 2026-02-20 |
| Scaling Verification Can Be More Effective than Scaling Policy Learning for Vision-Language-Action Alignment | [link](https://arxiv.org/abs/2602.12281) | ⚡ daily 2026-02-20 |

### 數據

| 论文 | 链接 | 备注 |
|:---|:---|:---|
| MIND: Benchmarking Memory Consistency and Action Control in World Models | [link](https://tldr.takara.ai/p/2602.08025) | ⚡ daily 2026-02-15 |
| EgoScale: Scaling Dexterous Manipulation with Diverse Egocentric Human Data | [link](https://arxiv.org/abs/2602.16710) | 🔧 daily 2026-02-20 |
| Benchmarking the Effects of Object Pose Estimation and Reconstruction on Robotic Grasping Success | [link](https://tldr.takara.ai/p/2602.17101) | 🔧 daily 2026-02-21 |
| SAGE: Scalable Agentic 3D Scene Generation for Embodied AI | [link](https://arxiv.org/abs/2602.10116) | 🔧 daily 2026-03-26 \| repo: https://github.com/NVlabs/sage |

### 部署

| 论文 | 链接 | 备注 |
|:---|:---|:---|
| Efficient Long-Horizon Vision-Language-Action Models via Static-Dynamic Disentanglement | [link](https://arxiv.org/abs/2602.03983) | 📖 daily 2026-02-10 |
| FreqPolicy: Efficient Flow-based Visuomotor Policy via Frequency Consistency | [link](https://arxiv.org/abs/2506.08822) | 🔧 daily 2026-02-20 |
| EdgeNav-QE: QLoRA Quantization and Dynamic Early Exit for LAM-based Navigation on Edge Devices | [link](https://arxiv.org/abs/2602.15836) | 🔧 daily 2026-02-20 |

### Sim2Real

| 论文 | 链接 | 备注 |
|:---|:---|:---|
| LAP: Language-Action Pre-Training Enables Zero-shot Cross-Embodiment Transfer | [link](https://arxiv.org/abs/2602.10556) | ⚡ daily 2026-02-13 |
| RADAR: Benchmarking Vision-Language-Action Generalization via Real-World Dynamics, Spatial-Physical Intelligence, and Autonomous Evaluation | [link](https://arxiv.org/abs/2602.10980) | ⚡ daily 2026-02-13 |
| Sample-Efficient Real-World Dexterous Policy Fine-Tuning via Action-Chunked Critics and Normalizing Flows | [link](https://tldr.takara.ai/p/2602.09580) | 🔧 daily 2026-02-15 |
| Sim2real Image Translation Enables Viewpoint-Robust Policies from Fixed-Camera Datasets | [link](https://arxiv.org/abs/2601.09605) | 📖 daily 2026-02-17 |
| Direction Matters: Learning Force Direction Enables Sim-to-Real Contact-Rich Manipulation | [link](https://arxiv.org/abs/2602.14174) | 📖 daily 2026-02-18 |
| Dex4D: Task-Agnostic Point Track Policy for Sim-to-Real Dexterous Manipulation | [link](https://arxiv.org/abs/2602.15828) | ⚡ daily 2026-02-19 |
| RoboGene: Boosting VLA Pre-training via Diversity-Driven Agentic Framework for Real-World Task Generation | [link](https://arxiv.org/abs/2602.16444) | ⚡ daily 2026-02-20 |

### 觸覺

| 论文 | 链接 | 备注 |
|:---|:---|:---|
| Semantic-Contact Fields for Category-Level Generalizable Tactile Tool Manipulation | [link](https://arxiv.org/abs/2602.13833) | 📖 daily 2026-02-18 |
| TactAlign: Human-to-Robot Policy Transfer via Tactile Alignment | [link](https://arxiv.org/abs/2602.13579) | 📖 daily 2026-02-18 |
| Symmetry-Aware Fusion of Vision and Tactile Sensing via Bilateral Force Priors for Robotic Manipulation | [link](https://arxiv.org/abs/2602.13689) | 📖 daily 2026-02-18 |

### 其他

| 论文 | 链接 | 备注 |
|:---|:---|:---|

| SCALE | [link](https://www.arxiv.org/abs/2602.04208) | 📖 daily 2026-02-11 |
| SceneSmith: Agentic Generation of Simulation-Ready Indoor Scenes | [link](https://arxiv.org/abs/2602.09153) | 📖 daily 2026-02-12 |
| NavDreamer: Video Models as Zero-Shot 3D Navigators | [link](https://arxiv.org/abs/2602.09765) | 📖 daily 2026-02-12 |
| Scaling World Model for Hierarchical Manipulation Policies | [link](https://arxiv.org/abs/2602.10983) | ⚡ daily 2026-02-13 |
| From Representational Complementarity to Dual Systems: Synergizing VLM and Vision-Only Backbones for End-to-End Driving | [link](https://arxiv.org/abs/2602.10719) | 📖 daily 2026-02-13 |
| Towards Learning a Generalizable 3D Scene Representation from 2D Observations | [link](https://arxiv.org/abs/2602.10943) | 📖 daily 2026-02-13 |
| The Observer Effect in World Models: Invasive Adaptation Corrupts Latent Physics | [link](https://tldr.takara.ai/p/2602.12218) | ⚡ daily 2026-02-14 |
| Agent World Model: Infinity Synthetic Environments for Agentic Reinforcement Learning | [link](https://tldr.takara.ai/p/2602.10090) | ⚡ daily 2026-02-16 |
| Real-to-Sim for Highly Cluttered Environments via Physics-Consistent Inter-Object Reasoning | [link](https://arxiv.org/abs/2602.12633) | 📖 daily 2026-02-17 |
| Scaling Single Human Demonstrations for Imitation Learning using Generative Foundational Models | [link](https://arxiv.org/abs/2602.12734) | 📖 daily 2026-02-17 |
| CLOT: Closed-Loop Global Motion Tracking for Whole-Body Humanoid Teleoperation | [link](https://arxiv.org/abs/2602.15060) | ⚡ daily 2026-02-19 |
| MeshMimic: Geometry-Aware Humanoid Motion Learning through 3D Scene Reconstruction | [link](https://arxiv.org/abs/2602.15733) | 🔧 daily 2026-02-19 |
| Feasibility-aware Imitation Learning from Observation with Multimodal Feedback | [link](https://arxiv.org/abs/2602.15351) | 🔧 daily 2026-02-19 |
| One Agent to Guide Them All: Empowering MLLMs for Vision-and-Language Navigation via Explicit World Representation | [link](https://arxiv.org/abs/2602.15400) | 📖 daily 2026-02-19 |
| Continual uncertainty learning | [link](https://tldr.takara.ai/p/2602.17174) | 🔧 daily 2026-02-21 |