Spatially Conditioned Diffusion Policy: Learning Precise and Robust Manipulation with a Single RGB Camera
Single-camera diffusion policy: points along intermediate end-effector trajectories in the denoising loop sample multi-scale image features, acting as attention anchors. Comparable to multi-camera baselines in sim.单相机扩散策略：在去噪循环中沿中间末端轨迹的点采样多尺度图像特征，作为视觉注意力锚点。仿真中表现与多相机基线相当。
BibTeX
@misc{kim2026spatially,
title = {{Spatially Conditioned Diffusion Policy: Learning Precise and Robust Manipulation with a Single RGB Camera}},
author = {Seoyoon Kim and Kanghyun Kim and Dongwoo Ko and Yeong Jin Heo and Min Jun Kim},
year = {2026},
eprint = {2606.14535},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.14535},
}