Adaptive Leakage-Aware Grasp Optimization for Flexible Granular Vacuum Grippers
Georg Siegemund, Frederik Hoeft, Arne Glodde, Franz Dietrich
Technical University Berlin, Institute of Machine Tools and Factory Management
11th CIRP Conference on Assembly Technologies and Systems (CATS 2026) — Procedia CIRP · 2026

Introduction
Flexible granular vacuum grippers can adapt to a wide range of object geometries, but their performance depends strongly on achieving an airtight seal. Leakage at edges, openings, textured surfaces, or during a misaligned approach can significantly reduce grasp reliability.
This work investigates whether a small number of pressure sensors integrated into the gripper can provide not only a global indication of leakage, but also information about where the leakage occurs.
At a glance
Conventional vacuum monitoring can detect insufficient vacuum but provides little information about the location or type of leakage.
Four pressure sensors positioned around the gripper membrane are compared against a calibrated baseline. Directional pressure deviations and global leakage magnitude are used to distinguish lateral from central leakage and reconstruct a spatial leakage field.
The experiments showed reliable differentiation between lateral and central leakage across the evaluated conditions. The reconstructed field provides information that can be translated directly into robot corrections.
Localising leakage instead of only detecting it
A conventional global pressure value can indicate that sealing is poor, but it cannot tell the controller whether the problem comes from lateral misalignment or insufficient contact in the centre of the gripper.
The distributed sensing concept used in this work relies on four internal pressure measurements to obtain spatial information without adding an external vision or tactile system.
From pressure signals to a leakage field
The processing pipeline moves from raw sensor readings to an actionable robot correction. Pressure readings from the four sensors are first corrected against a calibrated baseline, then converted into directional leakage gradients and an overall leakage magnitude.
This information is used to classify the leakage as lateral or central and to reconstruct a spatial leakage field, which is then translated into a concrete adjustment for the robot.
Using leakage information for robot adaptation
Directional, lateral leakage indicates that the end effector can be shifted in-plane to improve alignment with the object.
Central leakage indicates that additional contact force, more compliance, or an adjusted approach angle may be required instead.
Once the readings indicate a sealed condition, the gripper can proceed confidently with the final jamming phase that locks the granular medium in place.
Experimental evaluation
The approach was evaluated across different leakage configurations, suction settings, and loading conditions. The reconstructed leakage fields consistently reflected whether the leakage originated laterally or centrally, matching the intended failure mode in each test case.
Why it matters
The approach provides a lightweight sensing layer between simple global vacuum monitoring and much more complex tactile or vision systems.
It could ultimately support closed-loop grasp formation, in which the robot actively searches for a better seal before committing to the final grasp.
Highlights
- Four miniature pressure sensors on the inner rim of the membrane provide spatial leakage information without external vision or tactile hardware.
- A baseline learned across suction voltages turns the raw pressures into directional deviations and a global leakage magnitude.
- A hybrid reconstruction separates lateral from central leakage and produces a spatial leakage field; no misclassification occurred in the tested dataset of more than 300 samples.
- Tested on three surfaces (aluminium, varnished wood, 3D-printed PLA) and with three added weights (50 g, 115 g, 400 g).
Methodology
- Hardware: a FORMHAND FH-R80 granular vacuum gripper on a UR10e robot, with four pressure measurement points along the inner rim of the membrane at (R, 0), (0, R), (−R, 0), and (0, −R). The approach speed was limited to 20 mm/s.
- Baseline: the sealed, leak-free pressure response of the four sensors is recorded across suction voltages and interpolated linearly into a continuous baseline p0(V).
- Directional deviations: during operation each reading is compared with the baseline, Δp = p − p0(V), and converted into a normalised leakage weight between 0 and 1.
- Global leakage magnitude: the mean measured pressure is compared with the sealed reference pressure at full suction voltage (10 V), giving a value from 0 for a well-sealed contact toward 1 when the membrane fails to build up vacuum.
- Hybrid reconstruction: for an asymmetric pressure pattern, Gaussian radial basis functions at the sensor positions are blended by their leakage weights into a directional field. For a nearly symmetric pattern (low standard deviation across the sensors), a radial Gaussian at the centre, scaled by the global magnitude, represents central leakage.
- Data acquisition: Beckhoff EM3712 differential-pressure terminals (±1 bar, 4 ms sampling time) in a 1,000 Hz system, streamed to Python over TwinCAT ADS.
- Test conditions: five leakage scenarios (sealed, 10% and 25% lateral, 10% and 25% central), suction voltages from 0.2 V to 8.0 V in 0.2 V steps, three surfaces, and added weights of 50 g, 115 g, and 400 g. Lateral leakage was created by shifting the gripper 1–3 mm sideways, central leakage by lifting the membrane slightly or inserting a 0.5–1 mm spacer. More than 300 samples were collected.
Findings
- Lateral leakage: one sensor, or at most two adjacent ones, read a clearly higher pressure than the others. The difference to the sealed side was typically 40–100 mbar at 10% leakage and could exceed 150–200 mbar at 25%.
- Central leakage: the global pressure stayed high (poor vacuum) while the sensors differed little; the standard deviation across the four sensors stayed below 5 mbar.
- The reconstructed fields reflected the introduced leakage: rim-centred directional lobes for lateral leakage and symmetric central fields for central leakage. No misclassification occurred in the tested dataset.
- The global leakage magnitude rose with leakage severity at 6 V: for central leakage from 0.77 at 10% to 0.94 at 25%, and for lateral leakage from 0.59 to 0.80.
- Results were stable across all three surfaces, including the porous wood plate and the structured 3D-printed plate.
- Higher added weights improved sealing. Lateral leakage stayed clearly detectable under preload; central leakage became subtler but was still classified correctly from the low sensor variance.
Figures

Method diagram

Limitations
- The hybrid model assumes either symmetric or asymmetric leakage. Multiple simultaneous leak points, irregular object geometry, or sliding during the approach will need richer estimation.
- The directional resolution is limited to four sectors, which the paper found sufficient for robot adaptation.
- The central leakage prior is static and the lateral field uses smooth interpolation; neither fully captures nonlinear flow inside the membrane cavity.
- The evaluation used planar plates under controlled leakage conditions. Closed-loop robot adaptation based on the leakage field is described as a strategy and listed as future work, as is validation on diverse object geometries and dynamic grasping.
Citation
Siegemund, G., Hoeft, F., Glodde, A., & Dietrich, F. (2026). Adaptive Leakage-Aware Grasp Optimization for Flexible Granular Vacuum Grippers. Procedia CIRP — 11th CIRP Conference on Assembly Technologies and Systems (CATS 2026).
Interested in a research collaboration?
Get in touch to discuss joint research, data collection, or applied evaluation of new methods.