Skip to content
VLA 線 · 查看同日 AI 報告 →查看同日 AI 报告 →
VLA 研究日報 Pulsar
FEED
— AI 線暫無可用資料 —— AI 线暂无可用资料 —

VLA 研究日報VLA 研究日报

共 27 篇

🔧 技術技术

VLA [University of Southern California]

BIND: Binding 3D Robot Actions to 2D Image Features

Cameron Smith et al. · 提出把3D机器人动作绑定到对应2D图像特征的新动作表示,在物体位置与相机视角OOD下显著提升鲁棒性并大幅提高数据效率。可直接替换现有VLA动作头,本周即可在自有数据上验证。

VLA [Computer Science Department, University of Maryland, College Park]

What to Attend, What to Keep: Skill-Conditioned Visuotactile Representation with Progress-Guided Event Memory

Amir-Hossein Shahidzadeh et al. · 提出技能条件化的视触觉表征与进度引导事件记忆,让策略按阶段动态权衡视觉引导与触觉时序历史,直接服务抓取、对齐与接触判断。触觉VLA方向稀缺工作,可复用到现有视触觉策略。

📖 背景閱讀背景阅读

VLA [Amazon FAR (Frontier AI & Robotics)]

SimEX: Simulation-Integrated Robotics AutoResearch

Jiaheng Hu et al. · 让LLM编码agent在仿真中自动完成机器人研究闭环,把数字世界的自主科研能力延伸到物理世界。属自动化研究工具链,与VLA策略本身无直接接口。

VLA [The University of Melbourne]

The Planning Limits of Latent World Models

Ali Alrasheed et al. · 系统刻画潜在世界模型在何种预测条件下失效、何种条件下可用于规划,界定其能力边界。对世界模型驱动的VLA规划有认知价值,但非新方法。

VLA [XPENG Robotics]

IronMind: Scaling Humanoid Dexterous Manipulation via Camera-Space Ego-Centric Pretraining

Huimin Pan et al. · arXiv:2609.39403v1 Announce Type: new Abstract: Egocentric human video offers a scalable data source for dexterous manipulation, yet using it to train humanoid robots presents two challenges: (1) an embodiment gap, as human hands differ structurally from robot end-effectors and low-cost egocentric recordings lack the torso kinematics required by conventional retargeting; and (2) heterogeneous data quality, including noisy hand-pose tracking and weakly aligned text annotations. We introduce IronM