Humanoid robots promise general-purpose assistance, yet real-world humanoid loco-manipulation remains challenging because it requires whole-body stability, dexterous hands, and contact-aware perception under frequent contact changes. In this work, we study dexterous, contact-rich humanoid loco-manipulation. We first develop an RL-based whole-body controller that provides stable lower-body and torso execution during complex manipulation. Built on this controller, we develop a whole-body humanoid data collection system that combines VR-based teleoperation with human-to-humanoid motion mapping, enabling efficient collection of real-world demonstrations. We then propose Humanoid Transformer with Touch Dreaming (HTD), a multimodal encoder–decoder Transformer that models touch as a core modality alongside multi-view vision and proprioception. HTD is trained in a single stage with behavioral cloning augmented by touch dreaming: in addition to predicting action chunks, the policy predicts future hand-joint forces and future tactile latents, encouraging the shared Transformer trunk to learn contact-aware representations for dexterous interaction. Across five contact-rich tasks, Insert-T, Book Organization, Towel Folding, Cat Litter Scooping, and Tea Serving, HTD achieves a 90.9% relative improvement in average success rate over the stronger baseline. Ablation results further show that latent-space tactile prediction is more effective than raw tactile prediction, yielding a 30% relative gain in success rate. These results demonstrate that combining robust whole-body execution, scalable humanoid data collection, and predictive touch-centered learning enables versatile, high-dexterity humanoid manipulation in the real world.

Explore touch dreaming predictions interactively. The left panel shows the robot's head camera view. The right panel visualizes predicted vs. ground-truth touch signals—switch between Force, Latent Tactile, and Raw Tactile modes. For the latent tactile heatmaps, each latent dimension is independently normalized over the episode, with a minimum-range threshold derived from the most active dimension across all fingers to distinguish active from inactive latent contact regions. Note that this per-dimension normalization amplifies subtle changes and prediction errors in the latent space for better visibility.

@misc{niu2026htd,
      title={Learning Versatile Humanoid Manipulation with Touch Dreaming},
      author={Yaru Niu and Zhenlong Fang and Binghong Chen and Shuai Zhou and Revanth Senthilkumaran and Hao Zhang and Bingqing Chen and Chen Qiu and H. Eric Tseng and Jonathan Francis and Ding Zhao},
      year={2026},
      eprint={2604.13015},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2604.13015},
}