Human skin provides a rich tactile sensing stream, localizing intentional and unintentional contact events over a large and contoured region. Replicating these tactile sensing capabilities for dexterous robotic manipulation systems remains a longstanding challenge. In this work, we take a step towards this goal by introducing DexSkin. DexSkin is a soft, conformable capacitive electronic skin that enables sensitive, localized, and calibratable tactile sensing, and can be tailored to varying geometries. We demonstrate its efficacy for learning downstream robotic manipulation by sensorizing a pair of parallel jaw gripper fingers, providing tactile coverage across almost the entire finger surfaces. We empirically evaluate DexSkin's capabilities in learning challenging manipulation tasks that require sensing coverage across the entire surface of the fingers, such as reorienting objects in hand and wrapping elastic bands around boxes, in a learning-from-demonstration framework. We then show that, critically for data-driven approaches, DexSkin can be calibrated to enable model transfer across sensor instances, and demonstrate its applicability to online reinforcement learning on real robots. Our results highlight DexSkin's suitability and practicality for learning real-world, contact-rich manipulation.
DexSkin is an electrically addressable tactile sensor based on a capacitive mechanism. It is composed of a highly deformable dielectric layer sandwiched between two soft electrode plates. The electrode intersections form individually addressable taxels. We tailor the sensor into a single sheet that can envelop a curved, fingertip-like shape. On the right, we show one gripper finger assembly sensorized with DexSkin, composed of the sensing skin wrapped around a soft TPE sleeve and rigid PLA finger core.
High tactile spatial resolution is essential for tasks such as in-hand pose estimation. To evaluate DexSkin's resolution, the robot must pick up a pen and push it against a nearby cabinet to reorient it in-hand, using only tactile and proprioceptive feedback. Additionally, it must be robust to human perturbations at test time.
With DexSkin
Spatial Average Only
No Tactile
Tactile sensors are a promising source of signals for robots to learn by environment interaction. Here, we validate DexSkin's suitability for online learning by adapting a vision-only policy to perform delicate grasping of fragile fruits via real-world residual reinforcement learning.
Residual Policy trained with DexSkin
Random Residual Policy
No Tactile
Humans often use the dorsal side of their fingertips for tasks like flipping switches or removing batteries. To evaluate similar capabilities, the robot must secure the lid of a container with a rubber band. The robot is randomly provided with an intact band or a perforated band that will snap if used. In the latter case, the robot must retrieve a replacement.
With DexSkin, Perforated Band
No Tactile, Perforated Band
With DexSkin, Intact Band
No Tactile, Intact Band
Here we show visualizations of DexSkin tactile sensor readings while pressing on different regions of the finger with one or more contacts. These demonstrations highlight key attributes of DexSkin, such as spatial resolution, force sensitivity to light contacts < 2g, and coverage across the full finger surface.