日本フィジカルAI新聞

世界のフィジカルAIを、日本語で。

物体姿勢追跡arXiv:2605.23523

ComPose: 物体姿勢追跡における手の信頼タイミングの決定

ComPose: When to Trust Hands for Object Pose Tracking

シェア:XThreadsFacebookLINEはてブBluesky

RGB動画から手の動きを補助情報として活用し、重度の手による遮蔽下でも頑健に物体の6DoF姿勢追跡を行うフレームワークを提案した。

詳しい要約

1. どんなもの?

- 6DoF object tracking framework for hand-aware object pose estimation from RGB video. - Designed for embodied AI and robot manipulation. - Harmonizes hand motions as complementary cue for object tracking. - Combines object and hand cues from foundation models. - Adaptively selects informative hand joints. - Refines object motion using visible geometric evidence and learned correction. - Enforces temporal consistency over rotation and translation. - Yields stable 3D object trajectories without e…

2. 先行研究と比べてどこがすごい?

- Prior approaches rely heavily on strong external priors like depth data or 3D templates. - Remain highly vulnerable to severe occlusions by hand grasps despite explicit masks. - ComPose treats hand not purely as occluder but as complementary cue. - Eliminates need for external priors and explicit masks. - Achieves robustness under severe hand occlusion and geometric ambiguity. - Provides stable trajectories without external smoothing.

3. 技術・手法の肝は?

- Unified tracking pipeline combining object and hand cues from foundation models. - Adaptively selects informative hand joints. - Combines object- and hand-derived cues for motion estimation. - Refines object motion using visible geometric evidence and learned correction. - Enforces temporal consistency over both rotation and translation. - No external smoothing required.

4. どうやって有効だと検証した?

- Extensive experiments show accuracy, efficiency, and robustness under severe hand occlusion and geometric ambiguity. - Trajectories effectively transfer to downstream robot manipulation by enabling robots to reconstruct human actions from online videos.

5. 議論はある?

- 要旨からは不明

6. 次に読むべき論文は?

- 要旨で参照/比較されている研究は明示されていない。 - 関連手法として、depth-based object pose tracking, template-based tracking, hand occlusion handling, foundation models for pose estimation などが挙げられる。

※ AIが要旨から生成した要約です。正確性は原文をご確認ください。

著者: Jisu Shin, Junoh Lee, JunGyu Lee, Inhwan Bae, Dohyeon Lee, Hokyun Im, Youngwoon Lee, Hae-Gon Jeon

分類: cs.CV

原文アブストラクト

Reconstructing the motion of objects from videos is a key component for embodied AI and robot manipulation. While diverse approaches to object pose tracking have been studied, they rely heavily on strong external priors, such as depth data or 3D templates, and remain highly vulnerable to severe occlusions by hand grasps despite the use of explicit masks. In this work, we present ComPose, a 6DoF object tracking framework designed for hand-aware object pose estimation from RGB video. Rather than treating the hand purely as an occluder, our method harmonizes hand motions as a \textit{complementary cue} for object tracking. In detail, we recover a variety of object motions over time by combining object and hand cues from foundation models within a unified tracking pipeline. Here, ComPose adaptively selects informative hand joints, combines object- and hand-derived cues for motion estimation, and refines the resulting object motion using visible geometric evidence and a learned correction. We further enforce the temporal consistency over both rotation and translation, yielding stable 3D object trajectories over time without any external smoothing. Extensive experiments show that our method is accurate, efficient, and robust under severe hand occlusion and geometric ambiguity. In addition, the resulting trajectories can also effectively transfer to downstream robot manipulation by enabling robots to reconstruct human actions from online videos.

関連論文

  1. 1Boston Dynamics、車輪付きAtlasの研究試作機を紹介研究・技術
  2. 2GXOとApptronik、Apolloの物流現場試験を拡大 最大1,000台の展開を構想量産・展開
  3. 3newmoとKDDIが資本業務提携 2028年に堺市で自動運転タクシー商用化目指す提携・買収
  4. 4NVIDIA、CoRLでGR00T 2とIsaac Lab 3.0を発表新製品・発表
  5. 5UC BerkeleyとNVIDIAの研究者、AIエージェントでロボット制御コードを仮想環境で改善研究・技術

PR・アフィリエイト広告

FRONTIER(GPU搭載BTOパソコン)

FRONTIER(GPU搭載BTOパソコン)の公式広告バナーFRONTIERでGPU搭載PCを見る ↗
PR本紙発行元 EmplifAI