arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.11934cs.RO
收藏

Robot-DIFT: 用于接触丰富机器人操作的对应敏感扩散特征

Robot-DIFT: Correspondence-Sensitive Diffusion Features for Contact-Rich Robot Manipulation

TU Darmstadt(图宾根大学);KIT(卡尔斯鲁厄理工学院);FZI(弗劳恩霍夫研究所);Hessian.AI(黑森人工智能公司);Robotics Institute Germany(德国机器人研究所);Honda Research Institute Europe GmbH(本田欧洲研究院)

Yu Deng, Yufeng Jin, Xiaogang Jia, Jiahong Xue, Gerhard Neumann, Georgia Chalvatzaki

AI总结:

提出Robot-DIFT,通过流形蒸馏将扩散模型转化为确定性学生网络,结合空间-语义特征金字塔网络,为接触敏感任务提供实时对应敏感特征,在多个基准上超越现有方法。

摘要:

机器人操作常常在最后几毫米失败:策略可能识别出正确的物体,但忽略了动作所需的姿态偏移、边界或预接触对齐。我们认为,当语义不变性抑制了闭环控制的对应线索,或者这些线索未以可用形式暴露给策略时,就会发生此类失败。现代视觉编码器提供强大的语义抽象,但接触丰富的操作需要对应敏感性:对动作相关的姿态、边界和接触几何变化具有判别性特征响应。扩散特征为密集对应提供了强大的先验,但由于随机性、延迟和表示漂移,直接使用不切实际。我们引入了Robot-DIFT,一种用于实时控制的确定性扩散派生骨干网络。通过流形蒸馏,Robot-DIFT将噪声条件扩散教师网络转换为干净输入的单次学生网络,同时保留教师的特征流形。空间-语义特征金字塔网络(S2-FPN)将粗到细的学生解码器特征融合为视觉标记,向策略暴露语义上下文和精细接触细节。在RoboCasa、LIBERO-10和真实机器人上,Robot-DIFT在接触敏感任务上优于视觉-语言、自监督、几何导向和扩散基线。受控的骨干/读出交换表明,S2-FPN解锁而非取代了扩散对应先验。

英文摘要:

Robot manipulation often fails in the final millimeters: a policy may recognize the right object yet miss the pose offsets, boundaries, or pre-contact alignments needed for action. We argue that such failures arise when semantic invariance suppresses correspondence cues for closed-loop control, or when these cues are not exposed to the policy in a usable form. Modern visual encoders provide strong semantic abstractions, but contact-rich manipulation requires correspondence sensitivity: discriminative feature responses to action-relevant changes in pose, boundary, and contact geometry. Diffusion features provide a strong prior for dense correspondence, but direct use is impractical due to stochasticity, latency, and representation drift. We introduce Robot-DIFT, a deterministic diffusion-derived backbone for real-time control. Through Manifold Distillation, Robot-DIFT converts a noise-conditioned diffusion Teacher into a clean-input, single-pass Student while preserving the teacher's feature manifold. A Spatial--Semantic Feature Pyramid Network (S2-FPN) fuses coarse-to-fine Student decoder features into visual tokens that expose semantic context and fine contact detail to the policy. Across RoboCasa, LIBERO-10, and real robots, Robot-DIFT outperforms vision--language, self-supervised, geometry-oriented, and diffusion baselines on contact-sensitive tasks. Controlled backbone/readout swaps show that S2-FPN unlocks, rather than replaces, the diffusion correspondence prior.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑