FORTE: Tactile Force and Slip Sensing on Compliant Fingers for Delicate Manipulation

TL;DR

FORTE addresses delicate robotic manipulation by integrating tactile force sensing and slip detection into compliant fin-ray fingers via embedded fluidic innervation. The system provides high-rate sensing at $2\,\mathrm{kHz}$, enabling force estimation with RMSE $=0.187\,\mathrm{N}$ over $0$ to $8\,\mathrm{N}$ and slip detection latency under $100\,\mathrm{ms}$. It demonstrates a grasp success of $91.9\%$ across $31$ object types and a slip-detection F1 score of $0.91$ with precision $=1.0$, highlighting robust performance on fragile, slippery, and everyday objects. The approach benefits from easy 3D-printed fabrication and real-time tactile feedback, though temperature sensitivity of the sealed air channels points to future design refinements for broader robustness.

Abstract

Handling fragile objects remains a major challenge for robotic manipulation. Tactile sensing and soft robotics can improve delicate object handling, but typically involve high integration complexity or slow response times. We address these issues through FORTE, an easy-to-fabricate tactile sensing system. FORTE uses 3D-printed fin-ray grippers with internal air channels to provide low-latency force and slip feedback. This feedback allows us to apply just enough force to grasp objects without damaging them. We accurately estimate grasping forces from 0-8 N with an average error of 0.2 N, and detect slip events within 100 ms of occurring. FORTE can grasp a wide range of slippery, fragile, and deformable objects, including raspberries and potato chips with 92% success and achieves 93% accuracy in detecting slip events. These results highlight FORTE's potential as a robust solution for delicate robotic manipulation. Project page: https://merge-lab.github.io/FORTE/

Figures (5)

• Figure 1: Overview. FORTE enables the delicate manipulation of fragile objects by leveraging passive compliance and tactile feedback. This is achieved by integrating compliant fin-ray fingers and embedded fluidic sensors with algorithms for force estimation and slip detection.
• Figure 2: Hardware Design of FORTE. Fin-ray fingers with internal empty air channels are 3D printed as a single structure. The air channels act as a tactile sensor by measuring their pressure with an off-the-shelf pressure transducer. Each pair of channels (marked as the same color) under the inner and outer surfaces of the finger is sealed on the side surface and connected to one differential air pressure transducer.
• Figure 3: Grip Force Estimation Characterization (a) Snapshots of gripper in open and closed poses during data collection. Grip force of the fingers are measured with the horizontally placed loadcell testing rig. Scale bar represents 1 . (b) Sensor and load cell readings over time for a light and strong grip. Red and blue lines show sensor readings from the left and right fingers, respectively, with each line corresponding to a channel pair at distal, middle, or root sensing location. Sensor outputs exhibit a consistent monotonic trend with increasing grip force, with the distal sensors showing the largest response. This trend confirms that the pressure signals scale with finger deformation and support accurate force estimation. (c) Renders of the custom load-cell testing rigs. (d) Predicted force versus ground-truth force for models trained with the feature vectors from Section \ref{['sec:f_e']} as input. Colored dots represent data points collected using the corresponding testing rig, while the dashed black line indicates perfect prediction.
• Figure 4: Slip Detection Characterization(a) Snapshots of object displacements caused by stick-slip events. The time difference between the two snapshots of both objects is $66.7~ms$. The snapshots of apple corresponds to Slip Event 1. Scale bars represent 1 . (b--d) Filtered sensor readings, PSD features, and average moving variance of one representative trial of lifting an apple. There are four slip events during this trial. At the onset of each slip event, a sharp increase in the PSD feature is observed, resulting in a corresponding spike in the average moving variance, as further illustrated in the zoomed-in views.
• Figure 5: Experimental Setup (a) FORTE is mounted on a Franka Emika Panda robot arm for table-top grasping tests. Scale bar represents 5 . (b) Objects used for evaluation. Fragile objects are selected to evaluate force estimation accuracy and gripper compliance, while slippery items and eight everyday YCB objects are chosen to assess slip detection performance.