每日论文雷达

每日论文雷达

每天挑出值得机器人 / 具身智能研究者看的几篇论文，用大白话讲清楚。Type 1 是能直接拿来用的工作，比如可以当基线、可以借鉴某个模块；Type 2 是方法比较有潜力、可能改变现有做法的论文。

最新

11 篇 单日页
Type 1 · Directly applicable Requested 证据 B

Do As I Can, Not As I Say: Grounding Language in Robotic Affordances

LLM 给技能打「有用分」、价值函数打「能成分」，相乘选动作，真机长程指令执行成功 74%

Michael Ahn, Anthony Brohan, Noah Brown et al. Sichun Xu, Mengyuan Yan, Andy Zeng

收录 发布

Mobile ManipulationEmbodied ReasoningFoundation Models Project page

Type 2 · Promising method Requested 证据 B

Hierarchical World Models with Implicit Dynamics

把扩散噪声当高层动作搜子目标，长时程推T成功率从0.21升到0.77

Gaoyue Zhou, Yvonne Wu, Zichen Jeff Cui et al. Mido Assran, Lerrel Pinto, Yann LeCun

收录 发布

World ModelsRobot Learning Project page

Type 1 · Directly applicable Key group 证据 C

SPARK: Self-Correcting Embodied Memory via Physical Re-Observation

机器人自查记忆、飞回去重看再选择性改写，问答分从 0.31 升到 0.53

Dmytro Kurdydyk, Aadi Palnitkar, Sebastian Scherer, Seungchan Kim

收录 发布

Spatial MemoryScene Graphs

Type 1 · Directly applicable 证据 C

Probing Spatial State Representations in Visual Foundation Models

用冻结特征探针测十种视觉基础模型的空间状态：看得清≠记得住≠用得上

Baiqiao Yin, Zhongang Cai, Ruisi Wang, Qiwei Liang, Lei Yang, Ziwei Liu

收录 发布

Foundation ModelsSpatial MemoryWorld Models Project page

Type 1 · Directly applicable 证据 B

Learning to Live: A Step Toward Lifelong Embodied Intelligence

小模型流式门控决定何时记忆，平面图像记忆加自造工具，让机器人用一天经历答题

Lirong Che, Peidong Wang, Yimeng Su et al. Chuang Wang, Daqi Liu, Jian Su

收录 发布

Spatial MemoryAgentsEmbodied Reasoning

Type 2 · Promising method Key group 证据 C

Learning Agile Humanoid Dodgeball with Onboard Monocular Vision

G1 人形机器人只靠躯干单目相机躲球：用检测框历史估球位，三阶段课程训练全身闪避

Feng Tian, Yulin An, Chenhao Li, Matthias Heyrman, Marco Hutter

收录 发布

HumanoidsLocomotionSim-to-Real

10 篇 单日页
Type 1 · Directly applicable Must-readRequestedKey group 证据 B

StopWatch: Goal Grounding for Object-goal Navigation

物体导航常走到目标旁却不停；先让感知模型报出目标在否与距离，六个基准领先

Xinhao Liu, Yunsheng Tian, Yuheng Qiu et al. Xi Chen, Guanya Shi, Chen Feng

收录 发布

Object-Goal NavigationVLABenchmarks Project page

Type 1 · Directly applicable Must-readRequested 证据 A

Scaling Online Exploration for Robot Agents

冻结大模型的编程代理靠在线试错写出机器人程序，无示范在多数基准胜过示范训练策略

Runhan Huang, Xiangcheng Zhang, Haoyu Chen et al. Dawn Song, Weirui Ye, Yilun Du

收录 发布

AgentsRobot LearningMobile Manipulation Project page

Type 2 · Promising method Key group 证据 B

FLORA: An Open Model Suite for Scaling Robotics Data Engines

一套开源模型同时给机器人视频打分段字幕和奖励，成本约为顶级 API 的 1/125

Sicong Leng, Yunze Man, Zhiqi Li et al. Yuke Zhu, Johan Bjorck, Linxi "Jim" Fan

收录 发布

Robot LearningFoundation Models CodeProject page

Type 1 · Directly applicable Key group 证据 B

Learning Force-Reactive Humanoid Loco-Manipulation

手持测力夹爪采示教加力位混合全身控制，人形真机成功率80%对26.7%

Sihang Li, Xiaofeng Guo, Jinkun Cao et al. Chen Feng, Guanya Shi, Haozhi Qi

收录 发布

HumanoidsMobile ManipulationRobot Learning Project page

Type 1 · Directly applicable Key group 证据 B

EgoContact: Evaluating Physical Intelligence with Egocentric Video

用2.25万条人类第一视角接触标注，考模型“该碰物体哪里”，并把机器人抓取成功率翻倍

Adam Deryło, Alexey Gavryushin, Daniel Cremers, Marc Pollefeys, Xi Wang

收录 发布

BenchmarksRobot LearningFoundation Models Project page