Human-Centric Transferable Tactile Pre-Training for Dexterous Robotic Manipulation

At a glance

Problem: vision alone misses contact information needed for delicate manipulation. Source
Idea: share action and tactile representations across human pre-training and robot adaptation. Source
Data: H-Tac contains 160 hours spanning more than 300 tasks and 135,000 episodes. Source

Evaluation scope The real-robot evaluation separates ten in-distribution trials from five out-of-distribution trials per method and task. Source

TTP connects human tactile data collection, shared representations and downstream robot manipulation. Original figure

How it works

Map tactile readings onto a common hand representation to preserve spatial meaning across embodiments. Source
Train action and tactile experts together so the model predicts future contact as well as motion. Source

Separate understanding, action and tactile experts learn to connect observations with future motion and contact. Original figure

Evidence & results

On in-distribution fragile-chip placement, TTP succeeds in 80% of trials versus 60% without pre-training and 10% for BeingH-0.5. Evidence
In-distribution peeling with the Inspire hand reaches 23.33 cm on average, versus 14.65 cm without pre-training; this metric is length, not success rate. Evidence

Real-robot examples illustrate contact-sensitive tasks. Read the task-specific metrics alongside these selected demonstrations. Original figure

Limitations & boundaries

The small trial counts limit how strongly these results establish reliability beyond the tested conditions. Source
Plug insertion remains difficult: reported in-distribution success is only 20% with the gripper and 10% with DexBotic. Source

Takeaway

Tactile pre-training helps delicate tasks, but gains depend on the task and do not eliminate difficult contact failures. Source

Resources

Full text ↗PDF ↗

Figures: the paper's authors. Editorial source and image review: 2026-10-21. Findings are author-reported, not independently reproduced by Robotics Research Hub.

Original abstract

As an essential modality for dexterous and contact-rich tasks, tactile sensing provides precise force feedback that cannot be reliably inferred from vision. However, limited by hardware and data collection systems, existing datasets with tactility remain small in scale and narrow in contact coverage. Meanwhile, Vision-Language-Action (VLA) models with tactile modality are constrained on dynamics-agnostic post-training, which limits the performance ceiling on downstream tasks. In this paper, we present H-Tac, a large-scale tactile-action dataset with 160-hour egocentric human videos containing more than 300 tasks and 135k episodes. Building upon this, we propose Transferable Tactile Pre-Training (TTP), a system of tactile-based pre-training on human data for fine-grained robotic tasks. To bridge the gap between humans and robots, we use unified tactile and action spaces throughout the pre-training and post-training phases, preserving prior knowledge during human-to-robot transfer. By leveraging a tactile expert for future tactile prediction, our framework explicitly models the contact dynamics and precise physical interactions. Extensive experiments in simulation and on real robots demonstrate that our model achieves superior performance, exhibiting robust generalization and fine-grained manipulation capabilities. TTP paves the way for scalable tactile pre-training via human-to-robot transfer.

Citation