Robotiikan tutkimusartikkelit — sivu 6

AI7.10.

DepthWorld Builds Geometrically Consistent 3D World Models for Robot Manipulation

Current video world models miss geometric structure entirely.

Jai Bardhan, Josef Sivic, Vladimir Petrik

AI7.10.

Online Filtered Q-Gradients Speed Up Robot Flow Policy Training by 10x

QF3 uses filtered critic gradients to train flow-based robot policies, reporting zero-shot humanoid hardware transfer and 10× faster wall-clock training than FPO++.

Chung Min Kim, Brent Yi, David McAllister +7

AI6.10.

Transferable Adversarial Patches Disrupt World Action Models in Simulated Robot Tasks

We consider an attacker who aims to reduce the task success rate of a WAM-controlled robot by modifying a small region of its visual input.

Xuanyu Lu, Fengqing Jiang, Kaiyuan Zheng +8

AI6.10.

Recursive Video In-Context Learning Helps Robots Handle Long-Horizon Tasks

Wenrui Bao, Xinxin Liu, Bingxin Xu +1

Humanoid6.10.

How Humanoid Robots Grow Whole-Body Skills From Their Own Motion Data

Yucheng Zhang, Sirui Xu, Jinhong Li +8

Robotics5.10.

How Robots Navigate Active Sites as Work Zones and Hazards Change

CORNAV combines construction plans, work schedules, site perception, and safety rules to route robots around changing hazards.

Parastoo Ali Pour, Deepak Prakash Kumar, Tommy Zhou +2

AI5.10.

How Active Gaze Helps Robots Manipulate Precisely Without Wrist-Mounted Cameras

EyeRobot 2.0 is a robot learning framework that introduces Active Visual Fixation (AVF).

Kush Hari, Justin Kerr, Nidhya Shivakumar +7

AI5.10.

Geometric Learning from Novel Views with Lean Decoders and Strong Encoders

Keerthi Kaashyap, Dennis Anthony, Akshay Krishnan +5

Humanoid4.10.

Can Humanoid Robots Choose Tools and Complete Tasks While Moving Reliably?

Kyochul Jang, Seohyeon Park, Ohchul Kwon +9

Robotics4.10.

How GlassGuard Helps Mobile Robots Map Glass Without Blocking Safe Paths

GlassGuard combines camera detection, LiDAR geometry, and multi-view checks to help mobile robots map glass without blocking open paths.

Hanwen Guo, Zhengzhi Lin, Yusen Xie +1

AI4.10.

How DuoMind Helps Two Robots Coordinate With Shared Language

DuoMind helps two robots coordinate long manipulation tasks by sharing language-based plans, task information, and confidence between local controllers.

Hanchu Zhou, Dechen Gao, Hang Wang +5

AI3.10.

How Robots Learn Partner Limits for Safer Teamwork on New Tasks

Suyu Ye, Zheyuan Zhang, Vaishnav Tadiparthi +5

AI3.10.

FERPO Trains Robot Policies With Better Action Coverage, Faster Updates

Sebastian Sanokowski, Alireza Sarmadi, Majid Khadiv

AI3.10.

How RPG Helps Embodied Agents Improve Robot Skills Through Practice

RPG coordinates six specialized agents across its three stages: Constructor, Runtime Agent, Privileged Agent, Video Analyzer, Implementor, and Merger.

Yen-Jen Wang, Haozhe Jiang, Shuying Deng +7