日本フィジカルAI新聞

世界のフィジカルAIを、日本語で。

週刊ニュースレター購読
マニピュレーションarXiv:2611.01659

HuG: 能動知覚を備えたヒューマノイド全身アフォーダンス認識型把持の学習

HuG: Learning Humanoid Whole-Body Affordance-Aware Grasping with Active Perception

シェア:XThreadsFacebookLINEはてブBluesky

遮蔽された物体に対し視点選択と全身動作を統合し、制御器の安定性を考慮したシミュレーション学習のみで実世界87.5%の把持成功率を達成したフレームワーク。

詳しい要約

1. どんなもの？

- HuG は humanoid による whole-body affordance-aware grasping のためのフレームワーク - 対象は人間環境での物体把持 - 課題は egocentric 視点での部分遮蔽と、base 安定性を保ちつつ target を見続け到達する coordinated whole-body motion - Perceive-Align-Grasp pipeline で perception と action を結合 - simulation のみで学習し real demonstration はゼロ

2. 先行研究と比べてどこがすごい？

- 先行研究との具体的比較は要旨からは不明 - 特徴として viewpoint を明示的な decision variable として扱う belief-driven active perception を導入 - kinematic validity だけでなく controller-in-the-loop dynamics simulation でデータを選別 - kinematically valid な grasp の 73% を棄却し kinodynamically feasible な affordance を学習 - zero real demonstrations で real-world 87.5% 成功を報告

3. 技術・手法の肝は？

- Perceive-Align-Grasp pipeline の3段構成 - belief-driven active perception が occlusion を解消し viewpoint を決定変数として扱う - point-cloud-conditioned affordance policy が whole-body grasp を提案 - affordance-aware alignment が最も実行可能な configuration へ body を再配置 - floating base では kinematic validity が不十分との洞察 - controller-in-the-loop dynamics simulation で training data を curate

4. どうやって有効だと検証した？

- simulation のみで学習し real-world で評価 - multi-level shelf task で real-world success 87.5% - novel objects へ汎化 - 試験した各 height で少なくとも 73% の success - simulation の scale 実験で training 時に見ていない低い table height でも成功し空間的 extrapolation を示す

5. 議論はある？

- floating base における kinematic validity の限界を指摘 - kinematically valid でも deployment controller を不安定化する grasp を 73% 棄却 - kinodynamically filtered synthetic data が affordance を memorized configuration ではなく object geometry と whole-body feasibility に接地させるという主張 - 限界や失敗要因の詳細は要旨からは不明

6. 次に読むべき論文は？

- controller-in-the-loop dynamics simulation を用いた学習 - point-cloud-conditioned affordance policy - belief-driven active perception - whole-body grasping for humanoid - 具体的な参照論文は要旨からは不明

※ 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: Xuanbin Peng, Lai Wei, Yuze Zhai, Qingyuan Yu, Keyi Wang, Jianglong Ye, Xiaolong Wang

分類: cs.RO

原文アブストラクト

Acquiring objects in human environments is hard for humanoids: targets are partially occluded under egocentric viewpoints, and reaching them demands coordinated whole-body motion that keeps the target visible and reachable while the base stays stable. We present HuG, a framework for whole-body affordance-aware grasping with active perception that couples perception and action through a Perceive-Align-Grasp pipeline: belief-driven active perception resolves occlusion by treating viewpoint as an explicit decision variable; a point-cloud-conditioned affordance policy proposes whole-body grasps; and an affordance-aware alignment stage repositions the body to the most executable configuration. A key insight is that, on a floating base, kinematic validity is not enough: we curate training data with controller-in-the-loop dynamics simulation, rejecting 73% of kinematically valid grasps that would destabilize the deployment controller, so the policy learns kinodynamically feasible affordances. Trained entirely in simulation with zero real demonstrations, HuG attains 87.5% real-world success on a multi-level shelf task, generalizes to novel objects, and holds at least 73% success at every tested height. Beyond generalizing within its training distribution, the policy extrapolates spatially: at scale in simulation it succeeds at table heights well below any seen during training, indicating that kinodynamically filtered synthetic data grounds affordances in object geometry and whole-body feasibility rather than in memorized configurations.

arXivで原文を読む →arXiv概要ページ

関連論文

← 論文一覧
PR本紙発行元 EmplifAI