VLA论文精读目录

VLA Paper Taxonomy

Part 1：Foundation Model / Policy / Action Representation

1. Vision-Language-Action Foundation Models

这一类关注机器人基础模型：

\[(image, language) \rightarrow action \]

核心问题：

  • 如何将视觉语言模型迁移到机器人控制；
  • 如何利用大规模数据实现跨任务泛化；
  • 如何构建类似 GPT 的 embodied foundation model。

OpenVLA分类

  • Vision-Language-Action Foundation Model
  • Open-source VLA
  • Action Token Prediction

VLA 论文精读（一）OpenVLA: An Open-Source Vision-Language-Action Model
https://arxiv.org/pdf/2406.09246
GR00T N1分类

  • Humanoid Robot Foundation Model
  • Generalist Robot Policy
  • Embodied Foundation Model

VLA 论文精读（二）GR00T N1: An Open Foundation Model for Generalist Humanoid Robots
https://arxiv.org/pdf/2503.14734
π0分类

  • Vision-Language-Action Foundation Model
  • Flow Matching Policy
  • General Robot Control

VLA 论文精读（二十三）π0: A Vision-Language-Action Flow Model for General Robot Control
https://www.pi.website/download/pi0.pdf
π0.5分类

  • Open-world Generalization
  • VLA Scaling
  • Generalist Robot Policy

VLA 论文精读（十八）π0.5: a Vision-Language-Action Model with Open-World Generalization
https://arxiv.org/pdf/2504.16054
Gemini Robotics分类

  • Embodied Foundation Model
  • Vision-Language-Action
  • Large-scale Robot Intelligence

VLA 论文精读（二十五）Gemini Robotics: Bringing AI into the Physical World
https://arxiv.org/pdf/2503.20020
RDT-1B分类

  • Diffusion Foundation Model
  • Bimanual Manipulation
  • Large Robot Policy

VLA 论文精读（二十一）RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation
https://arxiv.org/pdf/2410.07864
SmolVLA分类

  • Efficient VLA
  • Small-scale Foundation Model
  • Affordable Robotics

VLA 论文精读（三十一）SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics
https://arxiv.org/pdf/2506.01844
VLA-0分类

  • VLA Construction
  • Zero Modification VLA
  • Foundation Model Analysis

VLA 论文精读（三十六）VLA-0: Building State-of-the-Art VLAs with Zero Modification
https://arxiv.org/pdf/2510.13054

2. Visuomotor Policy Learning and Action Generation

这一类关注机器人如何从 observation 生成 action。
主要路线：

  • Diffusion Policy
  • Flow Matching
  • Reactive Policy

Diffusion Policy分类

  • Visuomotor Policy
  • Diffusion-based Action Generation
  • Imitation Learning

VLA 论文精读（三）Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
https://arxiv.org/pdf/2303.04137
Reactive Diffusion Policy分类

  • Contact-rich Manipulation
  • Visual-Tactile Policy
  • Fast-Slow Control

VLA 论文精读（七）Reactive Diffusion Policy: Slow-Fast Visual-Tactile Policy Learning for Contact-Rich Manipulation
https://arxiv.org/pdf/2503.02881

3. Action Representation and Tokenization

核心问题：机器人动作是否可以像语言一样表示为 token。
Universal Actions for Enhanced Embodied Foundation Models分类

  • Action Representation
  • Universal Action Space
  • Embodied Foundation Model

VLA论文精读（十二）Universal Actions for Enhanced Embodied Foundation Models
https://arxiv.org/pdf/2501.10105
Survey on Vision-Language-Action Models: An Action Tokenization Perspective分类

  • Survey
  • Action Tokenization
  • VLA Taxonomy

VLA 论文精读（三十三）A Survey on Vision-Language-Action Models: An Action Tokenization Perspective
https://arxiv.org/pdf/2507.01925

Part 2：Data Scaling / Benchmark / 3D Representation / Dexterous Manipulation / Tactile

4. Robot Data Scaling and Simulation

这一类关注：机器人领域如何获得类似互联网规模的数据。
核心问题：

  • 真实机器人数据成本高；
  • simulation 是否可以扩展 robot learning；
  • 如何建立统一的数据平台和 benchmark。

DemoGen分类

  • Synthetic Demonstration Generation
  • Data-efficient Robot Learning
  • Simulation Data

VLA 论文精读（四）DemoGen: Synthetic Demonstration Generation for Data-Efficient Visuomotor Policy Learning
https://arxiv.org/pdf/2502.16932
ROBOVERSE分类

  • Robot Dataset Platform
  • Benchmark
  • Scalable Robot Learning

VLA论文精读（十一）ROBOVERSE: Towards a Unified Platform, Dataset and Benchmark for Scalable and Generalizable Robot Learning
https://roboverseorg.github.io/static/pdfs/roboverse.pdf
RoboTwin 2.0分类

  • Simulation Data Generation
  • Robot Benchmark
  • Domain Randomization

VLA 论文精读（二十八）RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization
https://arxiv.org/pdf/2506.18088
VR-Robo分类

  • Real-to-Sim-to-Real
  • Simulation Transfer
  • Robot Navigation and Locomotion

VLA 论文精读（二十九）VR-Robo: A Real-to-Sim-to-Real Framework for Visual Robot Navigation and Locomotion
https://arxiv.org/pdf/2502.01536

5. Benchmark and Evaluation

这一类关注：如何判断 VLA 是否真的具备泛化能力。
重点：

  • long-horizon tasks
  • language-conditioned manipulation
  • generalization

VLABench分类

  • Manipulation Benchmark
  • Long-horizon Reasoning
  • Language-conditioned Robotics

VLA 论文精读（五）VLABench: A Large-Scale Benchmark for Language-Conditioned Robotics Manipulation with Long-Horizon Reasoning Tasks
https://arxiv.org/pdf/2412.18194
ROBOVERSE分类

  • Benchmark Platform
  • Dataset Evaluation

VLA论文精读（十一）ROBOVERSE: Towards a Unified Platform, Dataset and Benchmark for Scalable and Generalizable Robot Learning
https://roboverseorg.github.io/static/pdfs/roboverse.pdf
RoboTwin 2.0分类

  • Simulation Benchmark
  • Manipulation Evaluation

VLA 论文精读（二十八）RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization
https://arxiv.org/pdf/2506.18088

6. 3D Representation and Geometric Understanding

这一类关注：VLA 是否需要显式三维世界理解。
核心问题：
RGB image:
↓
3D geometry:
↓
robot manipulation
PointVLA分类

  • 3D Vision-Language-Action
  • Point Cloud Representation
  • Spatial Understanding

VLA 论文精读（十四）PointVLA: Injecting the 3D World into Vision-Language-Action Models
https://arxiv.org/pdf/2503.07511
Lift3D Foundation Policy分类

  • 2D-to-3D Transfer
  • Foundation Policy
  • 3D Manipulation

VLA 论文精读（十五）Lift3D Foundation Policy: Lifting 2D Large-Scale Pretrained Models for Robust 3D Robotic Manipulation
https://arxiv.org/pdf/2411.18623
FP3分类

  • 3D Foundation Policy
  • Robotic Manipulation

VLA 论文精读（十六）FP3: A 3D Foundation Policy for Robotic Manipulation
https://arxiv.org/pdf/2503.08950

7. Affordance Learning

这一类关注：机器人不仅需要识别物体，还需要理解：物体可以如何被操作。
AffordDP分类

  • Affordance Representation
  • Generalizable Policy
  • Diffusion Policy

VLA 论文精读（十七）AffordDP: Generalizable Diffusion Policy with Transferable Affordance
https://arxiv.org/pdf/2412.03142

8. Dexterous Manipulation and Bimanual Control

这一类关注从arm + gripper走向复杂双臂、灵巧手、人类级操作。
ALOHA Unleashed分类

  • Bimanual Manipulation
  • Low-cost Hardware
  • Imitation Learning

VLA 论文精读（二十四）ALOHA Unleashed: A Simple Recipe for Robot Dexterity
https://arxiv.org/pdf/2410.13126
Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware分类

  • Bimanual Manipulation
  • Low-cost Robot Learning

VLA 论文精读（十九）Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware
https://arxiv.org/pdf/2304.13705
ManipTrans分类

  • Dexterous Manipulation Transfer
  • Residual Learning
  • Bimanual Control

VLA 论文精读（九）ManipTrans: Efficient Dexterous Bimanual Manipulation Transfer via Residual Learning
https://arxiv.org/pdf/2503.21860
DexGraspVLA分类

  • Dexterous Grasping
  • Vision-Language-Action
  • General Dexterous Manipulation

VLA论文精读（十）DexGraspVLA: A Vision-Language-Action Framework Towards General Dexterous Grasping
https://arxiv.org/pdf/2502.20900

Part 3：Tactile / Reasoning / Agentic Robotics / Fine-tuning / Surveys

9. Tactile Sensing and Contact-Rich Manipulation

这一类关注视觉模型在真实操作中的局限：

  • 接触状态不可见；
  • 物体形变难预测；
  • 力反馈缺失；
  • open-loop policy 容易失败。

核心方向：
vision
+
touch / force
+
feedback control
Beyond Sight分类

  • Heterogeneous Sensor Fusion
  • Vision-Language Policy
  • Multi-modal Robot Learning

VLA 论文精读（六）Beyond Sight: Finetuning Generalist Robot Policies with Heterogeneous Sensors via Language Grounding
https://arxiv.org/pdf/2501.04693
Reactive Diffusion Policy分类

  • Visual-Tactile Policy
  • Contact-rich Manipulation
  • Reactive Control

VLA 论文精读（七）Reactive Diffusion Policy: Slow-Fast Visual-Tactile Policy Learning for Contact-Rich Manipulation
https://arxiv.org/pdf/2503.02881
3D-ViTac分类

  • Vision-Tactile Learning
  • Fine-grained Manipulation
  • Sensor Fusion

VLA 论文精读（二十）3D-ViTac: Learning Fine-Grained Manipulation with Visuo-Tactile Sensing
https://arxiv.org/pdf/2410.24091

10. Reasoning, Planning and Agentic Robotics

这一类关注VLA 如何从observe → act，发展到observe → reason → plan → act → recover
CoT-VLA分类

  • Vision Chain-of-Thought
  • Visual Reasoning
  • Long-horizon Manipulation

VLA论文精读（十三）CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models
https://arxiv.org/pdf/2503.22020
RoboBrain分类

  • Robot Brain Architecture
  • Embodied Intelligence
  • Hierarchical Robot Reasoning

VLA 论文精读（二十七）RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete
https://arxiv.org/pdf/2502.21257
TIGeR分类

  • Tool-Integrated Reasoning
  • Geometric Reasoning
  • Vision-Language Model for Robotics

VLA论文精读（三十五）TIGeR: Tool-Integrated Geometric Reasoning in Vision-Language Models for Robotics
https://arxiv.org/pdf/2510.07181
FEAST分类

  • Robot Assistance
  • Personalization
  • In-the-wild Deployment

VLA 论文精读（二十六）FEAST: A Flexible Mealtime-Assistance System Towards In-the-Wild Personalization
https://arxiv.org/pdf/2506.14968
Multi-agent Embodied AI分类

  • Multi-agent Robotics
  • Collaborative Embodied Intelligence

VLA 论文精读（二十二）Multi-agent Embodied AI: Advances and Future Directions
https://arxiv.org/pdf/2505.05108

11. Fine-tuning, Adaptation and Efficient VLA

这一类关注大型 VLA 如何

  • 更快适配新任务；
  • 降低训练成本；
  • 提高推理速度。

Fine-Tuning Vision-Language-Action Models分类

  • VLA Fine-tuning
  • Training Strategy
  • Success Optimization

VLA 论文精读（八）Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success
https://arxiv.org/pdf/2502.19645
ConRFT分类

  • Reinforcement Fine-tuning
  • VLA Optimization
  • Consistency Policy

VLA 论文精读（三十二）ConRFT: A Reinforced Fine-tuning Method for VLA Models via Consistency Policy
https://arxiv.org/pdf/2502.05450
MoLe-VLA分类

  • Efficient VLA
  • Dynamic Inference
  • Layer Skipping

VLA 论文精读（三十）MoLe-VLA: Dynamic Layer-skipping Vision Language Action Model via Mixture-of-Layers for Efficient Robot Manipulation
https://arxiv.org/pdf/2503.20384
SmolVLA分类

  • Efficient Deployment
  • Small VLA Model

VLA 论文精读（三十一）SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics
https://arxiv.org/pdf/2506.01844

12. Surveys

这一类用于建立领域地图。
A Survey on Vision-Language-Action Models: An Action Tokenization Perspective分类

  • Survey
  • Action Representation

VLA 论文精读（三十三）A Survey on Vision-Language-Action Models: An Action Tokenization Perspective
https://arxiv.org/pdf/2507.01925
Pure Vision Language Action (VLA) Models: A Comprehensive Survey分类

  • Comprehensive Survey
  • VLA Taxonomy

VLA 论文精读（三十四）Pure Vision Language Action (VLA) Models: A Comprehensive Survey
https://arxiv.org/pdf/2509.19012
Efficient Vision-Language-Action Models for Embodied Manipulation: A Systematic Survey分类

  • Efficient VLA
  • Model Compression
  • Deployment

VLA论文精读（三十六）Efficient Vision-Language-Action Models for Embodied Manipulation: A Systematic Survey
https://arxiv.org/pdf/2510.17111

Research Questions

Data scaling

机器人是否可以像大语言模型一样，通过扩大数据规模获得能力提升？VLA 面临的核心问题是机器人数据昂贵、embodiment 不统一、动作空间不同以及长尾失败案例稀缺。主要路线包括：

  • Real Robot Data：真实机器人数据，例如 Open X-Embodiment、π0/π0.5。优势是物理真实性高，但规模有限。
  • Synthetic Data：例如 DemoGen、RoboTwin 2.0、ROBOVERSE。优势是可扩展，但存在 simulation gap。
  • Human Data：例如 teleoperation 和 egocentric video。优势是行为丰富，但人与机器人 embodiment 不一致。

Open Questions：

  • 是否存在 robot scaling law？
  • 什么样的数据最有价值？
  • failure data 是否比 successful demonstration 更重要？
  • 数据规模增加是否带来能力跃迁？

Representation

机器人需要什么样的世界表示？当前 VLA 主要依赖 image + language，但真实操作需要理解 geometry、object state、contact state、affordance 和 physical dynamics。主要方向包括：

  • Vision Representation：利用 CLIP、DINO、VLM 等视觉表示，语义理解强，但空间精度不足。
  • 3D Representation：例如 PointVLA、Lift3D、FP3，引入 depth、point cloud，提高空间理解能力。
  • Affordance Representation：例如 AffordDP，关注物体如何被操作，而不仅是物体类别。

Open Questions：

  • VLA 是否需要显式 3D 表示？
  • latent representation 是否已经包含足够空间信息？
  • robot memory 应该存储什么形式的状态？

Action generation

如何从视觉和语言生成机器人动作？

\[(o_t,l)\rightarrow a_t \]

主要路线包括：

  • Action Tokenization：例如 OpenVLA、Universal Actions，将连续动作转换为 token，使 Transformer 可以处理，但可能损失连续控制信息。
  • Diffusion / Flow Policy：例如 Diffusion Policy、π0、RDT-1B，直接生成动作轨迹，适合多模态和复杂 manipulation。
  • Reactive Policy：例如 Reactive Diffusion Policy，引入快速反馈，提高 contact-rich task 稳定性。

Open Questions：

  • action token 是否会成为机器人领域类似语言 token 的表示？
  • diffusion policy 是否适合作为通用控制接口？
  • policy 如何结合 reasoning 和 memory？

Memory

机器人如何利用过去经验？当前 VLA 通常学习：

\[a_t=\pi(o_t,l) \]

隐含假设当前 observation 足够，但真实任务具有长期依赖。
Memory 可以分为：

  • Short-term Memory：保存最近 observation/action，用于 temporal reasoning。
  • Episodic Memory：保存过去执行经验，例如类似任务的成功轨迹。
  • Semantic Memory：保存抽象知识，例如物体属性和操作规律。

相关方向：

  • RoboBrain；
  • Robot ICL；
  • MEM；
  • Hi Robot。

Open Questions：

  • memory 应该保存 observation、action 还是 latent state？
  • 如何进行 memory retrieval？
  • 如何避免错误经验影响策略？
  • memory 是否可以减少机器人数据需求？

Long-horizon planning

机器人如何完成多步骤、长时间任务？
例如：
prepare breakfast
↓
find bread
↓
open package
↓
place food
当前方法包括：

  • Language Planning：利用 LLM 分解任务。
  • Hierarchical Policy：高层 planner 负责任务规划，低层 policy 执行动作。
  • Reasoning-based Policy：例如 CoT-VLA，通过中间 reasoning 提升复杂任务能力。

Open Questions：

  • VLA 是否需要独立 planner？
  • skill library 是否仍然必要？
  • 长任务失败后如何恢复？
  • 如何判断任务完成？

Failure recovery

机器人失败后如何恢复？当前 imitation learning 主要学习成功轨迹，但真实环境包含 perception failure、action failure、planning failure 和 physical failure。
主要方向：

  • Feedback Control：加入 vision、force、tactile feedback。
  • Self-correction：机器人观察失败并修改策略。
  • Experience Replay：将失败案例加入训练或 memory。

Open Questions：

  • failure 是否应该成为主要训练数据？
  • 如何自动发现和分类失败？
  • 如何区分偶然失败和策略错误？
  • robot 是否需要类似 debugging 的能力？

Evaluation

如何评价机器人是否真正具备通用能力？传统 benchmark 主要关注 task success rate，但通用机器人还需要评价 generalization、robustness、adaptation 和 recovery ability。
当前方向：

  • Task Benchmark：VLABench、RoboTwin、ROBOVERSE。
  • Generalization Evaluation：测试 unseen object、environment、instruction。
  • Reliability Evaluation：测试失败率、一致性和恢复能力。

Open Questions：

  • success rate 是否足够？
  • 如何评价长期自主能力？
  • 如何评价 memory 和 adaptation？
  • 是否需要新的 robot intelligence benchmark？

Suggested Reading Path for Robot Research

Stage 1: Understand VLA Core

  1. Diffusion Policy
  2. OpenVLA
  3. π0
  4. π0.5

Stage 2: Understand Scaling

  1. GR00T N1
  2. Gemini Robotics
  3. RoboTwin 2.0
  4. ROBOVERSE

Stage 3: Understand Current Bottlenecks

  1. VLABench
  2. Reactive Diffusion Policy
  3. 3D-ViTac
  4. RoboBrain
  5. CoT-VLA

Stage 4: Explore Research Opportunities

  1. PointVLA
  2. Lift3D
  3. FP3
  4. DexGraspVLA
  5. RDT-1B
  6. TIGeR

CSDN按顺序

VLA 论文精读（一）OpenVLA: An Open-Source Vision-Language-Action Model
https://arxiv.org/pdf/2406.09246
VLA 论文精读（二）GR00T N1: An Open Foundation Model for Generalist Humanoid Robots
https://arxiv.org/pdf/2503.14734
VLA 论文精读（三）Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
https://arxiv.org/pdf/2303.04137
VLA 论文精读（四）DemoGen: Synthetic Demonstration Generation for Data-Efficient Visuomotor Policy Learning
https://arxiv.org/pdf/2502.16932
VLA 论文精读（五）VLABench: A Large-Scale Benchmark for Language-Conditioned Robotics Manipulation with Long-Horizon Reasoning Tasks
https://arxiv.org/pdf/2412.18194
VLA 论文精读（六）Beyond Sight: Finetuning Generalist Robot Policies with Heterogeneous Sensors via Language Grounding
https://arxiv.org/pdf/2501.04693
VLA 论文精读（七）Reactive Diffusion Policy: Slow-Fast Visual-Tactile Policy Learning for Contact-Rich Manipulation
https://arxiv.org/pdf/2503.02881
VLA 论文精读（八）Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success
https://arxiv.org/pdf/2502.19645
VLA 论文精读（九）ManipTrans: Efficient Dexterous Bimanual Manipulation Transfer via Residual Learning
https://arxiv.org/pdf/2503.21860
VLA论文精读（十）DexGraspVLA: A Vision-Language-Action Framework Towards General Dexterous Grasping
https://arxiv.org/pdf/2502.20900
VLA论文精读（十一）ROBOVERSE: Towards a Unified Platform, Dataset and Benchmark for Scalable and Generalizable Robot Learning
https://roboverseorg.github.io/static/pdfs/roboverse.pdf
VLA论文精读（十二）Universal Actions for Enhanced Embodied Foundation Models
https://arxiv.org/pdf/2501.10105
VLA论文精读（十三）CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models
https://arxiv.org/pdf/2503.22020
VLA论文精读（十四）PointVLA: Injecting the 3D World into Vision-Language-Action Models
https://arxiv.org/pdf/2503.07511
VLA 论文精读（十五）Lift3D Foundation Policy: Lifting 2D Large-Scale Pretrained Models for Robust 3D Robotic Manipulation
https://arxiv.org/pdf/2411.18623
VLA 论文精读（十六）FP3: A 3D Foundation Policy for Robotic Manipulation
https://arxiv.org/pdf/2503.08950
VLA 论文精读（十七）AffordDP: Generalizable Diffusion Policy with Transferable Affordance
https://arxiv.org/pdf/2412.03142
VLA 论文精读（十八）π0.5: a Vision-Language-Action Model with Open-World Generalization
https://arxiv.org/pdf/2504.16054
VLA 论文精读（十九）Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware
https://arxiv.org/pdf/2304.13705
VLA 论文精读（二十）3D-ViTac: Learning Fine-Grained Manipulation with Visuo-Tactile Sensing
https://arxiv.org/pdf/2410.24091
VLA 论文精读（二十一）RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation
https://arxiv.org/pdf/2410.07864
VLA 论文精读（二十二）Multi-agent Embodied AI: Advances and Future Directions
https://arxiv.org/pdf/2505.05108
VLA 论文精读（二十三）π0: A Vision-Language-Action Flow Model for General Robot Control
https://www.pi.website/download/pi0.pdf
VLA 论文精读（二十四）ALOHA Unleashed: A Simple Recipe for Robot Dexterity
https://arxiv.org/pdf/2410.13126
VLA 论文精读（二十五）Gemini Robotics: Bringing AI into the Physical World
https://arxiv.org/pdf/2503.20020
VLA 论文精读（二十六）FEAST: A Flexible Mealtime-Assistance System Towards In-the-Wild Personalization
https://arxiv.org/pdf/2506.14968
VLA 论文精读（二十七）RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete
https://arxiv.org/pdf/2502.21257
VLA 论文精读（二十八）RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization
https://arxiv.org/pdf/2506.18088
VLA 论文精读（二十九）VR-Robo: A Real-to-Sim-to-Real Framework for Visual Robot Navigation and Locomotion
https://arxiv.org/pdf/2502.01536
VLA 论文精读（三十）MoLe-VLA: Dynamic Layer-skipping Vision Language Action Model via Mixture-of-Layers for Efficient Robot Manipulation
https://arxiv.org/pdf/2503.20384
VLA 论文精读（三十一）SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics
https://arxiv.org/pdf/2506.01844
VLA 论文精读（三十二）ConRFT: A Reinforced Fine-tuning Method for VLA Models via Consistency Policy
https://arxiv.org/pdf/2502.05450
VLA 论文精读（三十三）A Survey on Vision-Language-Action Models: An Action Tokenization Perspective
https://arxiv.org/pdf/2507.01925
VLA 论文精读（三十四）Pure Vision Language Action (VLA) Models: A Comprehensive Survey
https://arxiv.org/pdf/2509.19012
VLA 论文精读（三十五）TIGeR: Tool-Integrated Geometric Reasoning in Vision-Language Models for Robotics
https://arxiv.org/pdf/2510.07181
VLA 论文精读（三十六）VLA-0: Building State-of-the-Art VLAs with Zero Modification
https://arxiv.org/pdf/2510.13054
VLA 论文精读（三十六）Efficient Vision-Language-Action Models for Embodied Manipulation: A Systematic Survey
https://arxiv.org/pdf/2510.17111

posted @ 2026-09-21 14:59  asandstar  阅读(11)  评论(0)    收藏  举报