Search papers, labs, and topics across Lattice.
BAIR
ETH
Tsinghua AIPKUShanghai InnovationShanghai Qi Zhi InstituteSharpaSJTUJun 11, 2026arXiv:2606.13102This paper introduces FTP-1, the first generalist foundation tactile policy capable of transferring tactile manipulation skills across diverse sensor setups and embodiments. By leveraging heterogeneous encoders and a shared tactile Transformer expert, FTP-1 was pretrained on extensive tactile manipulation data, enabling it to improve performance by 17.2% on known sensor configurations and achieve a remarkable 31% success rate on previously unseen sensors. This work establishes a unified baseline for tactile manipulation, paving the way for future advancements in tactile policy development.
FTP-1 not only excels on familiar tactile sensors but also achieves unprecedented success on unseen setups, redefining the potential for cross-sensor generalization in robotic manipulation.
Despite the success of vision-based generalist robotic policies, existing tactile-based policies remain tied to fixed embodiments and sensor setups. This is because tactile signals are highly heterogeneous across hardware, making cross-sensor generalization difficult. We present FTP-1,the first generalist foundation tactile policy pretrained to acquire transferable tactile manipulation abilities across diverse sensors and embodiments. FTP-1 supports varied tactile inputs, including image-, array-, and state-based signals, by using heterogeneous encoders to project them into unified morphology-aware latent tokens that are jointly modeled by a shared tactile Transformer expert. Pretrained on around 3,000 hours of tactile manipulation data aggregated from 26 data sources, spanning human and robot demonstrations across 21 sensors, FTP-1 learns tactile skills that transfer beyond the sensors seen during pretraining. Across downstream finetuning experiments spanning 5 hardware configurations, FTP-1 improves contact-rich manipulation on seen sensor setups by +17.2% and, surprisingly, transfers to two previously unseen tactile-sensor setups, achieving a +31% gain in success rate. FTP-1 establishes the first unified foundation baseline for tactile manipulation, providing future tactile policies with a shared model-level starting point. Pretrained models, datasets, training code and more visualization at https://ftp1-policy.github.io.
