AnySkin: Plug-and-play Skin Sensing for Robotic Touch

TL;DR

AnySkin tackles the barriers of using tactile sensing in robotics by decoupling the sensing interface from electronics and enabling replaceable, low-cost skins with improved signal consistency. The authors introduce fabrication changes, a self-aligning, self-adhering magnetic-elastomer skin, and a post-curing pulse-magnetized process to boost field strength. They demonstrate slip-detection capability and cross-instance generalization of learned policies, achieving zero-shot transfer across skin instances and favorable cross-sensor comparisons against ReSkin and DIGIT. The work shows that tactile data can be collected and reused at scale, enabling more robust visuotactile learning for precise manipulation, with practical benefits like rapid skin replacement. Limitations include environmental magnetic interference and the potential benefits of calibration or Faraday-style shielding for further improving robustness.

Abstract

While tactile sensing is widely accepted as an important and useful sensing modality, its use pales in comparison to other sensory modalities like vision and proprioception. AnySkin addresses the critical challenges that impede the use of tactile sensing -- versatility, replaceability, and data reusability. Building on the simplistic design of ReSkin, and decoupling the sensing electronics from the sensing interface, AnySkin simplifies integration making it as straightforward as putting on a phone case and connecting a charger. Furthermore, AnySkin is the first uncalibrated tactile-sensor with cross-instance generalizability of learned manipulation policies. To summarize, this work makes three key contributions: first, we introduce a streamlined fabrication process and a design tool for creating an adhesive-free, durable and easily replaceable magnetic tactile sensor; second, we characterize slip detection and policy learning with the AnySkin sensor; and third, we demonstrate zero-shot generalization of models trained on one instance of AnySkin to new instances, and compare it with popular existing tactile solutions like DIGIT and ReSkin. Videos of experiments, fabrication details and design files can be found on https://any-skin.github.io/

Figures (7)

• Figure 1: We present AnySkin, a skin sensor made for robotic touch that is easy to assemble, compatible with different robot end-effectors and generalizes to new skin instances. AnySkin senses contact through distortions in magnetic field generated by magnetized iron particles in the sensing surface. The flexible surface is physically separated from its electronics, which allows for easy replacability when damaged.
• Figure 2: (a) AnySkin is made by mixing Smooth-On DragonSkin 10 Slow and MQFP-15-7(25$\mu$m) magnetic particles in a 1:1:2 ratio, and curing it in the two-part molds shown above. Cured skins are magnetized using a pulse magnetizer. (b) Skins made with MQP-15-7(-80 mesh) and MQFP-15-7(25$\mu$m) particles. Note the concentration of particles at the surface of the former due to the larger particle size.
• Figure 3: Experimental setup used for slip detection experiments, where we train LSTM models on data collected by a Jaco Robot equipped with the AnySkin sensor (right). We train on a set of training objects (left top) and evaluate it on a set of unseen test objects (left bottom).
• Figure 4: Different circuit-skin alignments evaluated in Section \ref{['sec:alignment']}
• Figure 5: We evaluate the replaceability of AnySkin on a set of 3 precision tasks, where capturing the contact interaction using touch is critical (left). Our experimental setup consists of a Ufactory xArm 7 robot with an AnySkin sensor integrated into the standard gripper (center). Visual information is captured using three static cameras (1-3) and one egocentric camera (4) attached to the gripper. (right)