GelSlim 4.0

Tactile sensing provides robots with rich feedback during manipulation, enabling a host of perception and controls capabilities. Here, we present a new open-source, vision-based tactile sensor designed to promote reproducibility and accessibility across research and hobbyist communities. Building upon the GelSlim 3.0 sensor, our design features two key improvements: a simplified, modifiable finger structure and easily manufacturable lenses. To complement the hardware, we provide an open-source depth and shear field estimation algorithms to enable in-hand pose estimation, slip detection, and other manipulation tasks. Our sensor is accompanied by comprehensive manufacturing documentation, ensuring the design can be readily produced by users with varying levels of expertise. We validate the sensor’s reproducibility through extensive human usability testing.

The GelSlim 4.0 can either be mounted to a robotic gripper or used as a handheld sensor. The form factors for both the gripper sensors and handheld sensors share core components and use adapters to easily modify the sensor to fit your specific needs, such as increased rigidity.

All required 3D design files are available in both Solidworks and OnShape (coming soon). Files for laser cutting are provided in various formats, including original Adobe Illustrator files that are editable. With the exception of the OnShape files, which are cloud-based, all remaining files can be found in the GelSlim 4.0 Hardware GitHub repository. 

We provide adapters for the Franka Emika Panda Hand and the WSG-50 Gripper. We welcome community contributions for other grippers. 

We provide an adapter to convert the GelSlim 4.0 into a handheld sensor. This adapter keeps all the fragile ribbon cables on the inside of the sensor so that users can't twist and break them. This sensor is useful for expedited data collection, demonstrations with K-12 school groups, etc.

The GelSlim 3.0 was accompanied by an analytical depth estimation method based on photometric stereo and sensor calibration [Taylor et al. 2022]. This method required calibrating each new sensor. The GelSlim 3.0 used an expensive, optimized shaping lens to achieve the required illumination homogeneity of the contact surface to use this method. By refactoring the shaping lens, the GelSlim 4.0 removes a high-cost component at the expense of this analytical depth reconstruction method. Here, we present a learning-based approach for depth reconstruction in its place. In our approach, we train a U-Net [Ronneberger et al. 2015] to predict an RGB-to-depth mapping.

The optional dot pattern on the gel pads of the GelSlim 4.0 sensor enables optical flow-based tracking of dot displacements using standard OpenCV functions. These displacements are used to generate vector fields in pixel space. This shear field representation of tactile information has proven useful for force-rich manipulation tasks and slip detection [Zhang et al. 2018, Dong et al. 2020]. Our open-source repository for this algorithm can operate on either distorted or rectified sensor images. We offer three representations of shear in this implementation: the shear field, its resulting time-derivative, and curl and divergence decompositions. 

To contribute to GelSlim 4.0 algorithms, please use the pull request feature in the following repositories. To add a new repository to this list, please use our feedback form.

• Depth Estimation GitHub Repo

• Shear Field Estimation GitHub Repo

• How much does it cost to make a GelSlim 4.0 sensor?

Material cost per sensor is approximately $65. However, several materials must be purchased in larger quantities than you will use to make a single sensor, such as the silicone and silicone ink used to make the gel pad. This number also does not include tools like a laser cutter or 3D printer. Check out this spreadsheet for a full bill of materials.

• What's the difference between the gel pads with and without dots? 

The dots on the gel pad are used to easily estimate shear fields using optical flow. However, the dots also obscure the color gradients that give information about depth. Thus, shear field estimation methods that use these dots have less of the sensor area that can perceive color gradients. This is a trade-off that you should consider when deciding whether or not to use shear-tracking dots on your gel pads.