MemVLN: Streaming Episodic Memory for Vision-Language Navigation
给流式 VLM 导航加固定大小情景记忆，仅用官方数据 R2R 成功率 55%
Yan Wang, Jiangyong Huang, Siyuan Huang, Baoxiong Jia · · Vision-Language Navigation Spatial Memory
每日论文雷达
每天挑出最值得机器人 / 具身智能研究者看的几篇论文，用大白话讲清楚。Type 1 能直接用于手头项目，Type 2 可能改变研究方向。
今天
RoboHarness: A Simple Harness Outperforms VLA and World Action Models
不训练机器人数据，靠关键点跟踪和几何约束工具让大模型在仿真中胜过VLA
Bince Qu, Wei Chen, Bo Zhang · · Mobile Manipulation Agents VLA
RobotAgentBench: Towards Open-world Robot Autonomy
家居移动操作智能体基准：50 个布局、90 个原子任务，代码智能体+VLA 混合最强，开放场景仍不到 16 分
Lingjun Mao, Jinglin Cao, Lukun He, Wenpeng Xu, Yuchen Yan, Hengyu Jin 等 · · Benchmarks Mobile Manipulation Agents
SAFARI: Semantic- And Frontier-driven Autonomous Risk Inspection
四足机器人户外自主巡查危险物：把检测和「看见却没检出」都记进同一张风险地图
Ugur Kafkasyali, Mihnea Toader, Julia Richter, Turcan Tuna, Cesar Cadena, Marco Hutter · · Exploration Mapping Legged Robots
RAVEN: A Unified Active SLAM System for Rapid Autonomous Visual Exploration with Aerial Robots under Localization Uncertainty
四旋翼主动 SLAM：一个 λ 旋钮在探索快慢与定位精度间取舍，真机验证
Xinyi Chen, Yichen Zhang, Chenxing Jiang, Shaojie Shen · · Exploration Mapping Visual Localization
RESPLE2: Inertial-Informed Recursive Non-Uniform Spline Estimation for Multi-Sensor Odometry with Global Refinement
IMU 直接预测非均匀样条控制点，剧烈运动下多传感器里程计更准更稳
Ziyu Cao, William Talbot, Song Li, Pakpong Chirarattananon, Fredrik Gustafsson, Marco Hutter 等 · · Legged Robots Mapping
EgoContact: Evaluating Physical Intelligence with Egocentric Video
用2.25万条人手接触标注测模型“该碰哪”，换作检索记忆库后真机抓取成功率从16%升到58%
Adam Deryło, Alexey Gavryushin, Daniel Cremers, Marc Pollefeys, Xi Wang · · Benchmarks Robot Learning Mobile Manipulation
全部论文 7 篇
SAFARI: Semantic- And Frontier-driven Autonomous Risk Inspection
四足机器人户外自主巡查危险物：把检测和「看见却没检出」都记进同一张风险地图
Ugur Kafkasyali, Mihnea Toader, Julia Richter, Turcan Tuna, Cesar Cadena, Marco Hutter · · Exploration Mapping Legged Robots
EgoContact: Evaluating Physical Intelligence with Egocentric Video
用2.25万条人手接触标注测模型“该碰哪”，换作检索记忆库后真机抓取成功率从16%升到58%
Adam Deryło, Alexey Gavryushin, Daniel Cremers, Marc Pollefeys, Xi Wang · · Benchmarks Robot Learning Mobile Manipulation
RAVEN: A Unified Active SLAM System for Rapid Autonomous Visual Exploration with Aerial Robots under Localization Uncertainty
四旋翼主动 SLAM：一个 λ 旋钮在探索快慢与定位精度间取舍，真机验证
Xinyi Chen, Yichen Zhang, Chenxing Jiang, Shaojie Shen · · Exploration Mapping Visual Localization
RESPLE2: Inertial-Informed Recursive Non-Uniform Spline Estimation for Multi-Sensor Odometry with Global Refinement
IMU 直接预测非均匀样条控制点，剧烈运动下多传感器里程计更准更稳
Ziyu Cao, William Talbot, Song Li, Pakpong Chirarattananon, Fredrik Gustafsson, Marco Hutter 等 · · Legged Robots Mapping
RobotAgentBench: Towards Open-world Robot Autonomy
家居移动操作智能体基准：50 个布局、90 个原子任务，代码智能体+VLA 混合最强，开放场景仍不到 16 分
Lingjun Mao, Jinglin Cao, Lukun He, Wenpeng Xu, Yuchen Yan, Hengyu Jin 等 · · Benchmarks Mobile Manipulation Agents
MemVLN: Streaming Episodic Memory for Vision-Language Navigation
给流式 VLM 导航加固定大小情景记忆，仅用官方数据 R2R 成功率 55%
Yan Wang, Jiangyong Huang, Siyuan Huang, Baoxiong Jia · · Vision-Language Navigation Spatial Memory
RoboHarness: A Simple Harness Outperforms VLA and World Action Models
不训练机器人数据，靠关键点跟踪和几何约束工具让大模型在仿真中胜过VLA
Bince Qu, Wei Chen, Bo Zhang · · Mobile Manipulation Agents VLA
没有符合条件的论文。