每日论文雷达

Type 1 · Directly applicable Must-read 证据 B

Solv3R: 3D Looped Transformer as a Multi-view Solver

把多视图重建做成可接着算的学习型求解器，一个模型管离线、增量、重定位和流式

Jeong-gi Kwak, Andrea Tagliasacchi, Sho Kagami, Yuki Ono, Kwang Moo Yi

收录 发布

3D ReconstructionVisual LocalizationMapping CodeProject page

Type 1 · Directly applicable Key group 证据 C

VABAR: A Variational Automation Benchmark for Agentic Robotics

六类工业操作仿真基准，测编码智能体组技能的可靠性；人工方案也未满分

Shuangyu Xie, Kaiyuan Chen, Uksang Yoo et al. Yuke Zhu, Spencer Huang, Ken Goldberg

收录 发布

BenchmarksAgents

Type 2 · Promising method 证据 A

Perceptive Humanoid Motion from Simulation

只用仿真合成数据训练人形全身技能，零样本进6个陌生真实环境成功率91%

Siyuan Hu, Abulikemu Abuduweili, Yue Meng et al. Chen Chen, Matthias Müller, Vladlen Koltun

收录 发布

HumanoidsSim-to-RealRobot Learning Project page