Papers de cercetare

Rezumate în limbaj simplu ale celor mai recente cercetări în robotică — roboți umanoizi, roboți cvadrupedi, automatizare industrială și inteligență artificială.
499 lucrări
ChunkAct Lets Robots Reuse Task Plans Instead of Replanning Each Action
AIacum 26 de zile

ChunkAct Lets Robots Reuse Task Plans Instead of Replanning Each Action

ChunkAct compiles repeated robot procedures into reusable action chunks, reserving AI planning for key decisions and failures.

Bingxin Xu, Wenrui Bao, Xinxin Liu, Jiawei Zhou +2

A Mapless Flipper Controller Helps Tracked Robots Stay Level and Escape
Roboticsacum 27 de zile

A Mapless Flipper Controller Helps Tracked Robots Stay Level and Escape

A mapless flipper controller stabilizes an articulated tracked robot and helps it recover from stalls, including when tall grass hides a log.

Valentýn Číhala, Pavel Sarga, Martin Pecka +2

Robots Use Imagined Video to Build 3D Maps Before They Move
Roboticsacum 27 de zile

Robots Use Imagined Video to Build 3D Maps Before They Move

HorizonNav uses generated future views to build tentative 3D maps, then plans routes independently of the imagined camera motion.

Hajira Amjad, Valerii Serpiva, Viacheslav Liubimov +2

How Behavior-Tree Agents Make Onboard Robots More Reliable in Real Tasks
Roboticsacum 27 de zile

How Behavior-Tree Agents Make Onboard Robots More Reliable in Real Tasks

Fabio Hübel, Adam Lilja, Yue Yin +5

Thermal Vision Helps Robots Find Hidden Wheel Tracks in Snow Safely
Roboticsacum 28 de zile

Thermal Vision Helps Robots Find Hidden Wheel Tracks in Snow Safely

Yiming Yang, Jeremy P. Bos

Chronos Policy Uses Physical Time to Improve Robot Learning and Control
Roboticsacum 28 de zile

Chronos Policy Uses Physical Time to Improve Robot Learning and Control

Chronos Policy makes observation and action timing explicit in the policy.

Changhao Wang, Jaden Clark, Shuran Song

Robots Learn New Tasks by Changing the Noise Behind Their Actions
Roboticsacum 28 de zile

Robots Learn New Tasks by Changing the Noise Behind Their Actions

Noise Space Learning adapts a robot policy by changing its action noise while leaving the pretrained generator fixed.

Junjie Lu, Chuheng Zhang, Yucen Wang +4

How RoboRSI Helps Robots Improve and Reuse Skills in Real Environments
AIacum 29 de zile

How RoboRSI Helps Robots Improve and Reuse Skills in Real Environments

Zimo Wen, Yijin Chen, Yuxuan Cao +8

Fast Robot Trajectory Uncertainty Enables Safer Paths Through Crowds and Obstacles
Roboticsacum 29 de zile

Fast Robot Trajectory Uncertainty Enables Safer Paths Through Crowds and Obstacles

Zhiwei Xue, Jia Yue Kam, Jinhang Qiu +5

A Safety Filter Helps Humanoids Stay Safe Without Sacrificing Their Tasks
AIacum 29 de zile

A Safety Filter Helps Humanoids Stay Safe Without Sacrificing Their Tasks

FAITH separates task learning from a learned safety filter, helping robots preserve performance when safe actions exist and reduce harm when they do not.

Songyuan Zhang, Baljeet Singh, Sarthak Ranjeet Kaingade +2

Flexible Camera Policies Help Robots Handle Changing Views During Manipulation Tasks
Robotics10 oct.

Flexible Camera Policies Help Robots Handle Changing Views During Manipulation Tasks

VersaCamVLA helps robot manipulation policies adapt to changing camera arrangements using learned scene tokens and wrist-camera motion.

Boyao Han, Chen Shi, Jingjing Qian +2

Virtual Views Help Robots Improve Fine Manipulation Success at Test Time
Robotics10 oct.

Virtual Views Help Robots Improve Fine Manipulation Success at Test Time

Jiayu Wang, Yue Yu, Bin Zhu +2

A Balanced Data Diet Improves Large-Scale Reinforcement Learning for Robot Control
AI10 oct.

A Balanced Data Diet Improves Large-Scale Reinforcement Learning for Robot Control

Octi Zhang, Mateo Guaman Castro, Patrick Yin +4

How Context-Aware Safety Filters Keep Generated Humanoid Motions Safe Near People
AI9 oct.

How Context-Aware Safety Filters Keep Generated Humanoid Motions Safe Near People

CSF filters text-generated robot motions using natural-language safety rules and scene context, redirecting strikes when a person is present.

Lizhi Yang, Yiling Hou, Yao Tang +4

A Robot World Model That Tests Actions Before Robots Try Them
Robotics9 oct.

A Robot World Model That Tests Actions Before Robots Try Them

DreamTrue predicts robot-camera video after proposed actions and uses counterfactual training to improve action fidelity and policy evaluation.

Junyan Li, Ruizhi Li, Yu Liu +6

Learning Generalizable Dexterous Robot Manipulation from a Single Human Demonstration
Robotics9 oct.

Learning Generalizable Dexterous Robot Manipulation from a Single Human Demonstration

Jusuk Lee, Sungha Kim, Yeonsoo Park +8

Long-WAM Scales Visual Context for Predictive Robot Action and Control
AI8 oct.

Long-WAM Scales Visual Context for Predictive Robot Action and Control

Wei Huang, Bohan Zhang, Chenzhi Liu +13

Steer Robot Policies with Sparse Human Input Using RoboPrompt
Robotics8 oct.

Steer Robot Policies with Sparse Human Input Using RoboPrompt

Human intentions often underspecify robot actions.

Yanwen Zou, Chenyang Shi, Guoxuan Xu +5

Persistent 3D Object Memory for Spatial Questions in Egocentric Video
AI8 oct.

Persistent 3D Object Memory for Spatial Questions in Egocentric Video

We study how the objects retained, their estimated positions, movement histories, and contextual descriptions help answer different spatial questions.

Shravan Chaudhari, William Paul, Suchi Saria +2

When Is Robot Security Evidence Still Fast Enough to Act Safely?
Robotics7 oct.

When Is Robot Security Evidence Still Fast Enough to Act Safely?

Dimitrios Nikou, Nikolaos Kekatos, Sophia Petridou +1