PTLD: Sim-to-real Privileged Tactile Latent Distillation for Dexterous Manipulation
Learns tactile dexterous skills without tactile simulation: privileged sensors in the real world collect data to distil a tactile state estimator for a sim-trained policy. Reports 182% gain on in-hand rotation.无需触觉仿真即可学习触觉灵巧操作：在真实环境中借助特权传感器采集数据，蒸馏出基于触觉的状态估计器，用于仿真训练的策略。报告手内旋转任务提升 182%。
BibTeX
@misc{chen2026ptld,
title = {{PTLD: Sim-to-real Privileged Tactile Latent Distillation for Dexterous Manipulation}},
author = {Rosy Chen and Mustafa Mukadam and Michael Kaess and Tingfan Wu and Francois R Hogan and Jitendra Malik and Akash Sharma},
year = {2026},
eprint = {2603.04531},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2603.04531},
}Tactile & Visuotactile Sensing触觉与视触觉感知Dexterous hand灵巧手sim-to-realprivileged-learningin-hand-rotationdexterousreal